Changelog

Pipeline reliability & speed Single-source analysis now has recovery paths when a link yields useful evidence but no explicit thesis, and when the extractor labels a direct-source summary as ignored despite identifying the supplied URL as answering the request. Customers should see fewer empty or failed article-style and X reads when the provided source is factual, sparse, or not written as an argument.

Payments & metered billing Source-query payment challenges now honor the charge intent they advertise. Agents paying for a source read get a cleaner contract between the requested action, the wallet prompt, and the billable result.

Video, podcast & source analysis X post analysis now carries attached photos and videos through to the source record, with fallback handling for syndicated posts when native classification data is incomplete. Captured media bytes can also be archived durably, so visual source evidence is less dependent on remote URLs staying available.

Site, brand & public pages The know.fast homepage now says what it costs and what you get, up front: “Drop a link. Get cited knowledge back. For 1¢.”, a one-line packet description under the hero, and a note on why known links answer instantly for everyone at the same flat cent.

Research entities & personas The Steve Jobs persona gained deeper primary-material coverage across early interviews and authored letters. That makes the persona's answers less dependent on familiar summaries and more grounded in the source record behind the public figure.

Report quality & evidence fidelity Library recall now reaches the interactive research path: run chat can bring in related known sources as clearly labeled background, and agent recall can find semantically similar public and owned work before a new run is purchased. That makes follow-up questions and automated research less likely to start from scratch when Researcher already holds relevant evidence.

Pipeline reliability & speed Provided-source analysis now handles extraction-provider outages more gracefully, so a pasted article or link is less likely to collapse into a failed or empty run when one upstream fetch path is unavailable. Researcher can keep more customer-supplied evidence moving through the pipeline instead of treating transient provider trouble as the final result.

Site, brand & public pages Run and source pages now read brand identity from the same product bridge used by the rest of the public surface. Share cards, discovery metadata, and public handoff pages can stay consistent as Researcher, know.fast, and persona.energy evolve together.

Research entities & personas Persona and entity rebuilds now embed their searchable packets automatically after each index build. Freshly rebuilt expert corpora are ready for retrieval sooner, which makes persona answers and large-corpus browsing less dependent on a separate cleanup pass.

Report quality & evidence fidelity Report quality work moved upstream into the source library: shared retrieval, claim representation, ranking fusion, and new quality gates now give research runs a stronger evidence substrate before writing begins. The net effect is better recall for relevant prior knowledge and fewer thin reports when the answer depends on material Researcher has already seen. That prior knowledge now also reaches the interactive surfaces: run chat quietly recalls related sources the library already holds as clearly-labeled background (never as run evidence), and research-memory recall returns semantically related known public sources alongside your own stored work — so agents can notice existing coverage before paying for new research.

Video, podcast & source analysis Podcast episode chips now open show-level coverage pages, so a single Apple, Spotify, or Overcast episode can lead readers to the broader show catalog when it can be resolved. That makes podcast knowledge less isolated and turns episodes into entry points for discovering adjacent source material.

Agent interop & the contract Agent-facing discovery surfaces are now generated and guarded from the live tool catalog, contract versions, brand URLs, pricing, and trial policy. Agents should see more consistent MCP, package, and onboarding metadata because the public contract can no longer silently drift from the running service.

Site, brand & public pages Researcher, know.fast, and persona.energy now share explicit brand manifests for public identity, pricing language, trial policy, social metadata, and storefront routing. Customer-facing pages and cards keep the same voice today, but future brand or price changes have one source of truth instead of scattered copy.

Research entities & personas Persona groups gained their first durable shape: seeded group records and a panel chat endpoint make it possible to ask questions across a curated set of personas rather than treating each persona as an isolated surface. This starts turning persona collections into collaborative research panels.

Autonomous factory & operations Factory workflows were mirrored into the Auto agent surface, making Researcher operations more discoverable as named automation rather than hidden repo machinery. Brand-level metrics also now show runs, spend, and billed activity by product surface, giving operators a clearer read on where usage and cost are coming from.

Report quality & evidence fidelity User-provided source analysis now keeps the source's own evidence instead of narrowing too early around the request wording. That means a pasted article is less likely to produce a thin or citation-light answer just because the prompt was framed narrowly.

Pipeline reliability & speed Fresh single-article reads can now return an early knowledge packet while the deeper report continues, and the single-source path skips expensive cross-source passes that do not add value. New links feel faster and avoid wasting budget on work meant for multi-source research.

Payments & metered billing know.fast read counts now match the visible activity feed: every resolved read is counted as a know, while free and paid reads still keep their separate billing meaning. The page, tab, and activity table now tell the same story about who knows a source.

Video, podcast & source analysis Publisher coverage pages now show what a source itself declares, from feeds and sitemaps to video channels, and mark which items already have knowledge pages. Readers can move from one analyzed link to the wider body of available work without forcing every item through extraction upfront.

Agent interop & the contract Agent reads for fresh articles now get useful early packets instead of waiting for the full long-form path, with clearer title handling and status semantics. Source pages and read-on-click flows also agree more reliably about whether a link is already known or still needs its own knowledge page.

Site, brand & public pages know.fast run pages added a nearby copy-source action, publisher chips now open coverage pages, and the Knows view uses one consistent label and count. The Hermes-facing demo page also moved into a more deliberate public presentation with a character-led hero and cleaner brand language.

Payments & metered billing know.fast now separates genuinely billed reads from free prefix reads in the customer-visible signal, so agents and dashboards can tell when money actually moved. The new-user trial path also recovers cleanly instead of stranding someone after a failed or expired first attempt.

Video, podcast & source analysis Authenticated source access now runs through a registry of paywall credentials, so subscription-backed material can feed analysis without one-off wiring. Shared source pages also expose cleaner markdown for agent readers.

Agent interop & the contract Agent-facing know.fast calls now treat deferred links as active wait states rather than broken URLs, and the installable Hermes packages expose the same provider surface. Prefix-style reads also return clearer cost signals, so agents can branch on actual spend instead of guessing from a generic paid flag.

Chat & the run workspace Shared links in chat now render as compact previews, and persona chats read fresh X posts directly instead of failing behind a login wall. Mobile funding controls also fit the composer more cleanly, making paid follow-up less awkward on phones.

Site, brand & public pages Public surfaces got a practical polish pass: know.fast avoids broken-image flashes, X video labels are more accurate, entity report headings render cleanly, and persona pages keep consistent navigation through their proxied hosts.

Research entities & personas Persona coverage expanded with a new Jeremy Giffon configuration spanning interviews, X, and YouTube material, while public persona surfaces kept converging on clearer storefront-style presentation.

Payments & metered billing know.fast flat-fee reads now appear in Recent Transactions, so paid knowledge calls show up beside the rest of a customer's usage instead of hiding behind backend accounting. Persona marketplace listings also disclose both consult prices in the agent-facing payment shape before an agent pays.

Provider rails & failover Researcher now isolates X-related rate-limit pressure before it can slow unrelated customer work. When an external search rail is throttled, the system backs off instead of repeatedly retrying a constrained provider, keeping the rest of the queue healthier.

Agent interop & the contract know.fast is now easier to install and call from agents: package files expose a copy-ready prompt, the installable skill is the primary setup surface, and Hermes agents get both a Skills Hub package and a git-installable plugin. Fresh links can answer immediately with a deferred completion path, while existing share URLs resolve to their packet instead of asking the agent to re-analyze the same source.

Chat & the run workspace Run Analyst answers are better at absence claims: before saying a run does not mention something, the analyst now checks both the direct claim and the broader run evidence. That reduces false “not stated here” answers when the source material contains the idea in a different phrasing.

Site, brand & public pages know.fast pages and calls got smoother around first use: pasted-link visits can start directly on the knowledge page, the agent call-to-action is simpler, and article/X titles now use the recognizable source text instead of generic publication or handle labels. The public canvas also moved back to a cleaner white base with neutral hairlines.

Research entities & personas Paul Graham now has a dedicated public persona payment surface with marketplace-ready metadata and a standalone persona host, giving agents a concrete flagship persona to discover and pay for. A private Citrini entity also expanded persona coverage from a publication archive.

Payments & metered billing The flat-price knowledge call now presents a cleaner wallet-payment challenge and discovery shape, so automated callers can see the price, supported payment protocol, and know.fast realm before paying. Failed or duplicate reads continue to avoid unnecessary charges while successful paid reads reconcile back to the visible Knows count.

Video, podcast & source analysis Single-source analysis now exposes a richer source package on run pages, with extracted markdown, evidence, files, and podcast transcripts in one inspectable place. Podcast, X, article, GitHub, and video links also share more of the same resolution path, so previously known videos avoid repeat transcription and source previews use the material people recognize.

Agent interop & the contract know.fast is easier for a cold agent to discover and try: the first read can happen without an opt-in step, anonymous calls find the supported API, and the response surfaces source packages and stable status links instead of forcing agents to scrape a browser page. Persona and run chats use the same link-ingestion path, so a shared URL becomes grounded context consistently across chat surfaces.

Site, brand & public pages know.fast pages now lead with stronger source previews, including web-article imagery and a hero thumbnail overlay, and the Knows tab uses the same compact table language as the rest of the product. The signed-in activity view stays live with readable ages and a count that matches what the page shows, while Researcher-only Positions & ideas stay off the knowledge-page experience.

Video, podcast & source analysis Speaker attribution on video and podcast transcripts is now roster-first: the pipeline reads the naming evidence inside the audio itself — roll calls, introductions, people addressed by name — builds a cast list, and resolves every speaker label against it. Garbled spellings of the same person merge into one identity, a name that isn't supported by the audio never ships as if it were certain, and role labels like Chair stay honest when the audio never names the person. On by default for every run. Speaker attribution got a second, smarter pass: after the first labeling, the pipeline now reasons over the whole cast at once to recognize when one person was transcribed under several garbled names and unify them — while refusing to merge genuinely different speakers (verified it correctly kept two different council presenters apart). On by default. Speaker labeling on multi-speaker video/podcast is now several times faster. A redundant second labeling pass (added before the newer roster + reconciliation steps existed) was measured to add no quality while dominating latency — dropping it ran 3-7x faster with equal or better results across council meetings, a town hall, and an interview. Podcast and video links now resolve through one shared media-adapter path across Researcher and know.fast, including stronger Apple Podcasts handling, so first-time podcast links reliably become transcript-backed runs instead of depending on surface-specific routing.

Chat & the run workspace Run chat reads shared links too: paste an article, X post, or video link into a run's chat and the analyst ingests it via know.fast and responds — the same behavior as persona chat, now uniform across surfaces.

Site, brand & public pages know.fast's first-time processing state is calmer and clearer: new links now show a white, compact status row with a subtle live work signal instead of a gray block, including reduced-motion support.

Research entities & personas Share a link with a persona and it reads it: paste an article, X post, or YouTube/podcast link and ask what they think. Known links answer instantly from know.fast; a fresh video is transcribed in the background ('give me a minute'), then discussed.

Payments & metered billing Wallet payment sessions now open with enough deposit for a whole work-session of calls instead of exactly one, so agents stop paying channel-open overhead on every request — the deposit settles to what was actually spent and unspent funds are refunded. A used-up session now explains exactly how to close it and start a fresh one instead of failing opaquely, and failed analyses on flat-fee rails are never charged.

Agent interop & the contract know.fast now installs into Hermes agents in one step: pip install know-fast registers a know tool and a web-extract provider that speak the know.fast contract natively — instant cached knowledge packets, flat one-cent pricing, and stay-on-it polling for video and podcast transcription. The plugin now also carries payment challenges and returned agent keys through the setup flow, gives agents a free recovery path for deferred media runs, and caps waits so an extraction job cannot hang behind one long transcription. The agent contract itself also got a full UX overhaul: an agent's durable key now arrives in the response body of any paid call (shown once, at mint), deferred video and podcast runs can be polled through a public status link with no credentials, failures come back as a structured verdict that says whether a retry can work and whether any money moved (failed calls on flat-fee rails are never charged, and dead links answer instantly from cache for free), and text-only X posts now answer in seconds instead of waiting on the video pipeline. GitHub links now return the project's README with its real setup commands, and papers and PDFs carry their actual document titles. Installed agent clients now keep themselves current: every API response tells a client the latest published version of its own package and, when it is behind, the exact upgrade command to run — across know.fast, persona, and researcher.now clients. Version discovery docs now live at know.fast/.well-known and persona.energy/.well-known, and the nightly drift guard watches every published channel so no surface can silently go stale.

Site, brand & public pages know.fast signing in now stays where you are: login opens in place on every page, and the signed-in homepage becomes a simple dashboard of all your knowledge links with the paste box on top.

Site, brand & public pages The know.fast dashboard now shows only the links you actually ran through know.fast, in the familiar row format — thumbnail, title, the source link with one-click copy, and age.

Site, brand & public pages Cleaner know.fast knowledge pages: the Activity table gives the update text the room it needs (no per-event action column, tighter time), and the redundant Sources tab is gone since a knowledge page has a single source.

Site, brand & public pages know.fast knowledge pages now show a Reads feed instead of the internal pipeline log once a link is analyzed: when it was first known, and the running log of agents reading its knowledge since. While a link is still being analyzed, the live progress log stays.

Site, brand & public pages The know.fast Reads feed went live and complete: every read now counts — free browser reads, agents fetching the knowledge packet, and paid calls alike — and new reads stream onto the page the moment they happen, no refresh.

Pipeline reliability & speed Entity index rebuilds are now serialized and deduplicated: a burst of completing ingestion runs coalesces into one queued rebuild per persona instead of many concurrent full rebuilds. Interactive persona chats also get priority over bulk research and large-corpus style retrieval avoids full scans, so live answers stay responsive while background corpus work scales.

Provider rails & failover Production monitoring now probes the embeddings provider directly every few minutes, so an exhausted embeddings account pages immediately instead of surfacing hours later as degraded persona retrieval. Analyst answers and citation passes can also choose the fastest measured host for the same model, improving answer latency without changing the model customers receive.

Video, podcast & source analysis know.fast launched — a single-purpose surface where anyone (especially agents) sends a link and gets structured knowledge back fast: a live knowledge page for people, and a token-efficient markdown packet for agents with the summary and key claims inline plus pointers to the full text, graded evidence, and timestamped transcript. One POST in, one packet out, wallet-payable without an account.

Agent interop & the contract Analyze runs' markdown export became a compact knowledge packet built for agent relay, and branded API surfaces now rewrite every self-URL to the calling product's host — an agent talking to know.fast lives entirely in know.fast URLs. Personas also learned to read shared links: send any persona a URL in chat and it comments on the actual content, reusing stored transcripts for previously analyzed videos and fetching fresh articles live.

Chat & the run workspace Persona conversations now feel better tuned to the surface they happen on: Telegram answers are shorter by default, compact source links replace pasted URLs, and final prose is delivered before slower source cleanup finishes. Persona memory can carry useful context across sessions for the same customer, while group chats avoid personal memory and balance chatter.

Site, brand & public pages Persona.energy is cleaner as a public directory: the list refreshes quickly after new bots or personas ship, sorts between usage and A-Z, and keeps each row focused on the person and the chat action. Persona pages on the microsite now hide research-plumbing tabs, keeping the experience centered on asking the modeled person.

Research entities & personas Telegram feedback capture: when the team asks a user for feedback through a persona bot, the user's next reply is recorded as feedback and acknowledged — never consumed as a paid question. Operators send asks (or arm capture for one already sent) and read responses from one place.

Research entities & personas Personas now greet first-time visitors with starter questions: each persona generates ~6 corpus-grounded example questions from its own profile and report at build time, and every surface offers them — tappable chips on the persona page, inline buttons on the Telegram bots (/start, /help, and right after linking), and a starterQuestions field for agents on persona listings. On Telegram the buttons behave like a real product: a tapped question is echoed into the chat and the answer replies to it, the tapped row flips to a checkmark so it can't double-bill, a dice button pages through the full list in place, and each answer offers the persona's suggested follow-up as a one-tap next question.

Research entities & personas Winston Churchill and Aristotle joined the persona fleet, expanding Researcher beyond contemporary founder and investor voices into primary-source public-domain corpora. The builds added stronger curation for catalog stubs, misattributions, classic translations, and direct full-text ingestion, so historical personas can answer from original material instead of loose web summaries.

Autonomous factory & operations Groundwork landed for splitting heavy persona building onto a dedicated machine: workers can now be scoped to claim only interactive work or only build work, so customer conversations never compete with corpus ingestion for the same box. The metrics overview now answers operator questions directly — total platform sources, customer funds, users, signup activity, answer length, consult channel, and provider cost — while painting instantly with public data before admin-only sections resolve.

Site, brand & public pages Menus now open as bottom sheets on phones. The account menu, run filters, and per-run actions all slide up from the bottom of the screen with a dimmed backdrop instead of floating as small desktop-style popovers, making them easier to read and tap on mobile.

Research entities & personas The first Telegram persona bot canary is ready for Paul Graham: a Telegram chat or group can set up an existing Persona account through persona.energy, ask paid grounded persona questions with /ask or a bot mention, and receive final answers through the same durable persona chat and billing path as the web app. The setup link now makes the email-code step explicit, keeps both code-step actions full width, avoids Telegram webview zoom on input focus, and ends with a minimal "Done, go back to Telegram" status instead of an account-management card. Telegram persona bots also graduated from the single-bot canary to a registry: every persona can now get its own bot through one env-held config entry (Stanley Druckenmiller is the first addition), with per-bot webhooks, command menus, and chat buttons provisioned by a single script.

Payments & metered billing Per-query source reads now have wallet-native payment coverage, so an agent can pay to retrieve one evidence package without opening a full research run. Persona chat funding also refreshes balances correctly after deposit, making low-balance recovery feel immediate instead of stuck.

Agent interop & the contract Agents can now open Researcher run, persona, and source links as native markdown, and source pages can return the full evidence package as structured data. That gives agent clients a supported way to read cleaned text, transcripts, citations, and synthesis without scraping browser pages.

Site, brand & public pages Persona.energy became a fuller public entry point for the persona layer, with persona-branded navigation, legal links, social preview imagery, a calmer light default, and an interactive "How it works" explainer. The public shell now feels like a coherent product surface rather than a thin proxy.

Research entities & personas The persona experience now has its own standalone front door, installable agent path, and richer visual explanation of how persona evidence moves through Researcher. Public persona pages also present cleaner navigation and social previews, making persona corpora easier to share with both people and agents.

Autonomous factory & operations The operations metrics overview is easier to scan: recent billing transactions now use the full dashboard width, so long run titles and customer identifiers do not squeeze the table into a narrow column.

Video, podcast & source analysis Paper and PDF links now route into single-source analysis the same way videos, podcasts, articles, and posts do. Dropping an arXiv, preprint, or PDF URL into the composer starts from that document instead of treating it like a broad web crawl.

Chat & the run workspace The source library is now a live workspace instead of a static explorer: recent sources load from production data, filter by type, search by title or origin, and page through older evidence without bogging down. The same research composer sits at the top, so browsing prior evidence and starting the next question now happen in one place.

Research entities & personas Persona coverage expanded with Jessica Livingston and ultimape, both with recognizable profile imagery, while entity tables shed low-value type labels so the dashboard focuses on the person and their evidence. Source lists also use cleaner origin labels and platform icons, making large persona corpora easier to scan.

Pipeline reliability & speed Source selection is more fault-tolerant: if a small retry batch cannot produce a structured model answer, Researcher keeps the strongest ranked candidates for the final merge instead of failing the run. That means a shaky triage step can still preserve useful evidence and complete the research path.

Pipeline reliability & speed Source-heavy runs and account pages are faster to revisit: reports load before lower-priority activity work, previews are prepared ahead of time, and very large persona updates avoid more memory limits. Extraction recovery also keeps delayed provider results and retryable outages from turning into misleading missing-source states.

Payments & metered billing Persona chat now shows the cost of a message before the user sends it and turns an empty balance into a clear deposit action instead of a failed composer. Wallet-native agent payments now cover article analysis and expert persona consults as well as research runs, settling before billable work begins and keeping paid requests tied to the wallet that funded them.

Provider rails & failover Video discovery now uses multiple providers instead of depending on one YouTube path, with extra hardening when YouTube throttles transcript or metadata lookups. Provider outages also keep clearer retry and failure boundaries, so customers are less likely to see a thin source set with no explanation.

Video, podcast & source analysis Single-source pages are now full web-app surfaces: they use the same width as the run page, show website bodies inline when available, and provide direct actions to open or copy the source. Media discovery is also more resilient because Researcher can fall back across providers when one video source is quota-limited or throttled.

Agent interop & the contract Wallet-native agents can now pay directly for article analysis and persona consults, not only deep research runs. Those calls return durable ownership and credentials after payment, so an agent can fund a single task and continue through the supported Researcher contract.

Chat & the run workspace Chat now handles more real-world edges: the mobile sheet stays clear of the keyboard, the live cue simply says when Researcher is writing, and streamed answers are no longer replaced by a weaker cleanup pass. Underfunded persona chats also switch to a deposit action instead of leaving the user in a composer that cannot send.

Site, brand & public pages The signed-in dashboard is easier to scan as an account grows: filters now cover source type, text search, and Research-versus-Analysis work, with custom inline dropdowns that match the rest of Researcher. Deposit dialogs also got a calmer layout, keeping payment actions clearer without pulling the page out of proportion.

Research entities & personas Persona surfaces are calmer and more trustworthy for paid consults: the page shows per-message cost, offers a deposit path when balance is low, and keeps the answer the user watched stream instead of swapping it out mid-turn. Entity profile summaries also drop more internal indexing labels, keeping the public profile focused on readable source context.

Provider rails & failover Source extraction now preserves provider and run-control results that settle just after a per-source timeout. Source-only ingestion can still record a failed source record for an outage, while ordinary research runs keep propagating retryable run-control failures.

Video, podcast & source analysis X post media now carries through as source thumbnails, including backfilled previews for earlier runs. Source-heavy pages can show the image that customers recognize from the original post instead of a generic card.

Chat & the run workspace Chat streaming finished its consolidation across run, job, entity, topic, and global chat surfaces. Answers now use one steadier live-update path, making reloads, completions, and follow-up turns behave more consistently across the workspace.

Research entities & personas Persona coverage expanded with Peter Thiel and Fred Wilson corpora, including essays, blogs, X archives, interviews, and recognizable profile imagery. Entity evidence also filters transcript evidence to the named speaker by default, so persona retrieval is less likely to blend in interviewer or guest voices.

Payments & metered billing Wallet-native pay-per-call is live: an agent can fund work directly from a stablecoin wallet with no account, and each call settles on-chain — priced so every call stays net-positive.

Agent interop & the contract Agents can now onboard with no human and no signup: a paid call with no key returns a payment challenge, the wallet pays, and the response hands back a durable API key — now over MCP as well as HTTP. The skill is also published on npm (@researcher-now/skill) and PyPI (researcher-now) so agent runtimes can install it in one step.

Site, brand & public pages Published a full Terms of Service and Privacy Policy with a plain acceptance notice, stating clearly that Researcher is experimental and provided as-is.

Report quality & evidence fidelity Reports are easier to verify in place: grouped citations open a full evidence drawer, hover previews stay readable for longer source lists, and live run pages begin filling the Sources tab before the run finishes. Discovery also got more selective, so date-sensitive prompts and authority-heavy topics are less likely to be pulled off course by weak or misleading sources.

Pipeline reliability & speed Runs now surface more of their work as it happens and avoid more silent failure modes: active runs can show sources before completion, streamed answers no longer downgrade back to waiting states, and extraction problems are reported instead of quietly dropping evidence. Large source and persona collections also keep paginating instead of hitting hidden caps.

Provider rails & failover Provider failures are more visible at the evidence boundary: when extraction providers are unavailable, Researcher reports the outage instead of making it look like sources simply disappeared. That gives customers a clearer reason for gaps and a better path to retry or continue later.

Video, podcast & source analysis Single-source link runs now have better visual previews, including website screenshot fallbacks when a source does not provide a usable lead image. Source pages also scale better for large corpora, so individual articles, posts, or media sources remain reachable even when a persona has thousands of items.

Agent interop & the contract Agent-facing surfaces got sturdier: topic pages now include a ready-to-copy agent handoff, and the published agent packages gained fuller tool coverage so external agents can inspect Researcher work through the supported contract instead of scraping pages.

Chat & the run workspace The chat workspace is harder to strand: silent streams now time out, streamed answers no longer fall back to an older waiting state, and mobile sheets keep the composer clear of the browser toolbar. Recents also bring cross-platform persona sessions into one list, making prior conversations easier to resume.

Standing topics & monitoring Topic watches can now reuse shared search and RSS registry results while paginating through longer X result sets, reducing duplicated collection work as a topic grows. Topic pages also gained a cleaner prompt/status layout and an agent handoff action, making a watch easier to inspect or pass to another agent.

Site, brand & public pages Researcher now has a site-wide System/Light/Dark theme with a true dark surface, cleaner account controls, and polish across public pages, dashboards, metrics, and mobile tables. The signed-in dashboard can page through older runs, so past work is reachable without relying on a short initial list.

Site, brand & public pages Runs and the dashboard now show a real preview image instead of a blank tile: a link run uses the page's own lead image, falling back to a live screenshot of the page, and a run anchored on a single post surfaces that post's attached photo or video. Text-only posts with nothing to show keep a clean lettered placeholder.

Research entities & personas Persona and source-scale browsing got stronger: larger authored archives can flow through the standard persona builder, Casey Handmer now has a recognizable profile image, and large persona source pages no longer fall over at old row limits. Persona chat also gained a unified cross-platform Recents list, making prior persona conversations easier to pick back up.

Report quality & evidence fidelity Clicking a grouped citation chip now opens that citation's sources in the same panel as the end-of-report Sources list — a sidebar on desktop, a bottom sheet on phones — so you can read every cited site, title, and excerpt for a claim without losing your place. Hovering still gives a quick peek — and that peek now stays put while you move into it and scrolls through long source lists instead of slipping away.

Pipeline reliability & speed Long runs now preserve more completed work when something interrupts them: deploys drain active jobs before restarting, stalled sources are time-bounded, provider outages are detected sooner, and resumed synthesis reuses finished report and merge passes instead of starting those expensive steps over.

Agent interop & the contract The agent distribution loop is now active instead of advisory: the daily surface guard checks the website, skill copies, npm package, Hermes plugin, public tool docs, and published channels, then republishes stale committed-main artifacts before opening an issue. Hermes also exposes the same source-package retrieval path as MCP, so agent clients can inspect full source evidence without falling out of the supported contract.

Chat & the run workspace Chats recover faster after a reload because the page keeps the live authorization state instead of waiting through repeated bridge checks before reconnecting. That makes in-progress replies and run-scoped conversations less likely to look stalled when the browser refreshes. The positions-and-ideas table also reads correctly on a phone again: it was being forced into the compact one-line row layout meant for the activity and sources lists, which crushed its narrow columns until short labels like "LONG" wrapped one letter per line. It now stacks each idea into a clean labeled card — instrument, side, thesis, evidence, and type — matching how the other report tables collapse on mobile.

Standing topics & monitoring Topic follow-up now stays fast as topics grow: saved evidence is searched through a per-topic semantic index with a bounded fallback, so source-grounded answers can start streaming without waiting on a large corpus scan. The topic page also reads more clearly: the prompt now sits flush beneath the status panel as one continuous card (matching the run page) with the action row below it, and a new "For agents" action copies a ready-to-paste handoff so another agent can follow the topic through Researcher's public tools.

Site, brand & public pages The account menu was consolidated into one calmer surface: starting a new run or a new topic now lives at the top of the menu, and the light/dark theme picker tucks into a collapsible row instead of sitting open all the time. The separate plus button was retired, so the top bar is just your balance and avatar.

Autonomous factory & operations Public changelog narration now runs as a daily batch instead of reacting to every push, so maintenance stays current without noisy automation churn. Those changelog commits also use a deployment-recognized author, keeping the public site update path reliable.

Report quality & evidence fidelity Reports now make evidence easier to inspect without distracting from the read: citations are grouped more quietly, source chips lead with recognizable site icons, and the Sources drawer exposes the underlying material when you need it. The writer also became more prose-first around tables and can carry source media and image galleries into the report, so visual evidence is available without turning the brief into a dump of raw artifacts.

Pipeline reliability & speed When many topics watch the same source — a shared search or feed — Researcher now fetches it once and reuses the result across topics rather than paying for the same lookup repeatedly, so monitoring stays efficient as more topics come online. Background turns also recover more cleanly after restarts or stale running states, reducing the chance that customer work appears stuck after deploys or worker interruptions.

Payments & metered billing Standing topics now meter the real cost of every background scan and show running spend right on the topic page. The cost of keeping a topic current is metered and billed back to its owner at the same rate as research runs, batched into periodic charges rather than per-scan, and a topic whose balance runs short pauses until it is funded again instead of quietly accruing.

Video, podcast & source analysis Media and source ingestion are now on by default for more research paths: Google Doc seeds can enter the corpus directly, podcasts can reuse existing transcripts, and media-heavy runs only show a missing-transcript state when the run is actually centered on media. That makes mixed-source research feel less brittle while keeping the run viewer focused on the evidence type that matters.

Chat & the run workspace Run, entity, and topic conversations now share steadier streaming behavior and a clearer action bar, so follow-up questions can move between asking and agent-oriented actions without each surface inventing its own controls. Chat status also reconciles against the authoritative run state, reducing false failures after a reply actually completed.

Chat & the run workspace The chat dock now looks the same on every surface: run, persona, and topic chats share one flat panel instead of the run chat carrying its own drop shadow, so moving between them feels like a single consistent workspace. The general chat composer also shows the same scope chip as the in-page chats — when a conversation is anchored to a specific run you can see that context and clear it to drop back to a general chat in one click.

Standing topics & monitoring Standing topics became a real monitoring workspace: clean handle-based URLs, shared-source reuse, visible scan spend, and balance-aware pauses make them easier to share, scale, and pay for honestly. Topic owners can now ask private, source-grounded follow-up questions from the topic page, with streamed answers drawn from the topic's saved evidence.

Standing topics & monitoring Asking a topic a follow-up question now returns in seconds instead of sometimes taking a couple of minutes: a topic's saved evidence is indexed for fast semantic search and the lookup is time-bounded, so grounded answers stream back quickly and fall back gracefully rather than stalling on a large topic.

Site, brand & public pages Topics now live at clean, shareable handle-based addresses — your topic reads as a name-and-topic path rather than an account-settings URL — laying the groundwork for public topic profiles. Existing links keep working and redirect to the new form.

Research entities & personas Prior persona chats are reachable again: a sessions expander returned to the persona chat header, so you can open the list of your earlier conversations with a persona and jump back into one whenever you want, while the chat still opens as a clean single column by default.

Research entities & personas Persona chat is sturdier: your messages no longer vanish after sending — a finished answer is kept even if the follow-up sync hiccups, and a failed turn now leaves your question in place with a clear error instead of wiping the exchange. Behind that, persona answers fail far less often: the model gets room to finish long replies, a real second model backs up the first, and when a structured answer can't be produced the chat now falls back to a plain-prose answer rather than erroring out.

Research entities & personas Persona chat feels alive while it works and can't silently hang: the wait now shows a rotating, persona-flavored status with a ticking timer instead of a frozen "Getting ready", the answer streams in smoothly character by character rather than in big sudden blocks, and a stalled generation is now caught and ended on both the server and the client so the chat can never get stuck mid-answer with no way out. The chat dock also lost its drop shadow for a cleaner page.

Research entities & personas Persona chat now survives a page reload mid-answer: if you refresh while a reply is still generating, the chat re-attaches to that live turn and keeps streaming it to completion instead of abandoning it. And a reply you watched stream in is no longer quietly replaced by a shorter one — when a follow-up citation check would otherwise regenerate a briefer answer, the longer answer you already saw is kept (with any stray citation cleaned up).

Provider rails & failover Provider outages now pause affected research runs and resume them after backoff instead of spending retries immediately or ending with a misleading failure. Customers are more likely to see work continue once credits, rate limits, or availability recover rather than having to restart the run.

Video, podcast & source analysis Source collections can now be enumerated from publisher-declared structures and compared with what Researcher already holds. That sets up clearer coverage views: which pages, videos, podcast episodes, or posts are already in the corpus, and which gaps still need ingestion.

Agent interop & the contract Agents can now pull a single source's full package directly over MCP: cleaned source text or transcript, graded evidence, synthesis, and citation handles back to the exact source and moment. Persona answers and standing topic briefs now preserve the source links and conversation memory agents need to move from a cited claim to the underlying evidence without oversized extraction blobs.

Chat & the run workspace The "Thinking" indicator now shimmers through a soft sunrise palette — pale blue warming through tan and peach to coral and pink, sweeping across the text instead of a flat grey pulse. It gives in-progress runs and replies a little more life while staying muted and on-brand, and it quietly steps aside for anyone who prefers reduced motion.

Chat & the run workspace Chat on a phone is dialed in: the chat opens as a bottom sheet you can swipe down to dismiss, the keyboard no longer zooms the page when you tap the box, the composer and buttons sit clear of the home indicator and notch, and tap targets are bigger. The same conversation works in the run chat and the persona chats.

Chat & the run workspace Run and persona chat now share a calmer, more durable workspace: softer motion, fade-in messages, clearer starter prompts, warm surfaces, and visible scope chips keep follow-up questions anchored to the report or expert in front of you. The chat window is also easier to leave and resume, with Escape support, focus returning to the launcher, and in-progress replies that feel active without reading as an error.

Site, brand & public pages The whole interface was warmed: surfaces and borders shifted from cool grey-white to a soft warm off-white across every page. It's a subtle change that makes Researcher read calmer and more considered without touching contrast or legibility.

Research entities & personas Persona chat got a cleaner, calmer header and a more reliable composer: the header now leads with the persona's avatar and name instead of a verbose title and source tally, the input bar sits flat without a drop shadow, and the send button no longer gets stuck inactive after your first message — while a persona is answering it becomes a stop control, so you can cancel and immediately ask a follow-up.

Agent interop & the contract Researcher is now self-describing when an agent is given the product domain: compatible clients can discover the MCP server and available research tools without a hand-copied endpoint. That makes the first connection path simpler for article analysis, deep research, and persona consults.

Report quality & evidence fidelity Run answer grounding got stricter about what actually improves accuracy: targeted retrieval for flagged finance facts was evaluated and kept out of the production path after it failed to rescue the remaining errors. The guard now also catches unsupported figures while an answer is being generated and gives the model one focused retry, so hard numbers are more likely to be grounded before the final cleanup pass.

Payments & metered billing Funded persona consults now use the same per-answer payment rails as wallet-native keyless consults, so every successful quick or deep answer is authorized, settled, and accounted for. Paid consults can also return immediately for slower async work and settle only after the answer succeeds, keeping expert persona calls responsive without charging failed answers.

Video, podcast & source analysis Analyze runs now ask for a direct, answer-first summary of what the source argues, with core claims, findings, figures, and takeaways cited to the source. Run pages also collapse duplicate run-level artifacts, so customers see the final brief and synthesis without repeated intermediate copies.

Agent interop & the contract Agents can consult expert personas and read standing topics as first-class tools: ask Paul Graham, Warren Buffett, Elon Musk, and more a question and get an answer in their voice with citations, or pull the living brief from any of your standing research topics. Standing topics also gained richer watched-source coverage, persona-guided interpretation, current cited claim graphs that can include durable, checkable social posts as opinion-tier evidence, collapsible brief sources, and durable reviews; the discovery contract now presents the service clearly as Researcher, exposes wallet-native price and payment protocol details for analysis, deep research, and persona consults, and tells calling agents how to list personas, pay without signup, reuse the returned key, recognize authenticated read routes as API-key calls, run keyed consults asynchronously at a fixed per-question price, recover gracefully from stale persona session handles, and surface durable viewer links. The install path is now a concrete agent setup page for Codex MCP, Hermes, wallet-native Tempo, and the first known-corpus ask; agents can call ask_corpus when they already know the public corpus slug instead of guessing whether to use persona chat or research memory.

Chat & the run workspace First-time start chats are more honest about whether a run actually exists: if a long prompt is classified as a run-summary request before any run has been created, Researcher now turns it into a normal runnable research proposal instead of leaving the customer with a successful-looking chat and no run in their account. Mobile run pages also keep Activity, Sources, and Artifacts as compact one-line rows with fresh relative timestamps visible on the right, making it faster to scan progress and open details on a phone.

Site, brand & public pages The homepage now brings stronger finished-run examples closer to the top, with video analysis, equity research, and consumer thesis work visible before the deeper capability tour, while topic detail pages give Activity its own tab. Daily briefs are easier to scan too: supporting sources sit behind a compact expandable control, thesis-delta changes render as calmer cards, and expanded sources use dense cited-source rows instead of a tall stack.

Research entities & personas Researcher added durable persona seeds for a founder corpus, a longevity evidence panel, and a subscription-publication voice, covering long-form founder material alongside geroscience research, clinical longevity frameworks, NAD+ skepticism, measured self-experimentation, and tech strategy analysis, while persona builds now enrich those profiles with full social archives, quota-resilient long-form video discovery, duration filtering, appearance checks that keep videos by the person instead of just about them, rebuilt reports, and recognizable public avatars by default, including missing portraits for the new longevity and strategy profiles. Persona chat also keeps malformed structured responses from leaking raw JSON, and public persona surfaces support a restorable unavailable state so the directory only shows personas that are ready.

Autonomous factory & operations Automated maintenance and recovery now use the same skeptical review standard as a hard gate before repair work can ship or recover a customer run, and that review now covers every real change rather than only traditional code edits. When that gate passes, approved tickets and repairs can move through merge, deploy, and live-release verification automatically, while repeated ticket updates no longer start duplicate shipping runs for the same request; the independent review path now authenticates through managed secret injection, and Factory activity rows now open a per-run detail page so customers can inspect what each automation event actually did.

Report quality & evidence fidelity Research quality now has a daily canary: lightweight extraction coverage and persona-fidelity checks can watch for regressions and only raise an issue when measured quality drops. This turns report faithfulness from a manual eval habit into an operating signal, and run answers now use a prove-or-remove audit so hard specifics need same-granularity retrieved evidence behind them before they can remain in customer-visible answers.

Pipeline reliability & speed Recoverable source-only run failures now trigger an immediate self-heal path instead of waiting for a periodic audit, with signed failed-run completion events reaching the repair loop reliably. The repair loop keeps regression testing, deploy locking, customer run recovery, and public verification tied together so failed runs can move back toward completion with less operator delay.

Payments & metered billing Run chat now builds its semantic retrieval index as part of the run and accounts for that embedding work in the run's metered cost. Customers get better follow-up answers without hidden, unpriced compute sitting outside the billing trail.

Agent interop & the contract Standing topic briefs now treat watched accounts as source candidates, not automatic signal: individual X posts are checked for topical relevance, source cards prefer items the narrative cited, and each scan saves useful items into a retrievable topic memory for stronger future briefs and topic answers. Daily briefs also respect the calendar-day boundary, and topic status now reflects the running watch state from creation while bootstrap outcomes are recorded as events instead of backend lifecycle states.

Site, brand & public pages Terms acceptance now stays out of the way entirely: anonymous visitors can read public pages without a compliance bar, and signed-in use records acceptance silently. Product surfaces stay focused on the work while the standing Terms and Privacy links remain available. The topics feed's table/cards switch also moved into the table header as quiet icons, so the list leads with content instead of a control. The homepage's example runs were refreshed to current work — a long video distilled to a read, an equity deep-dive, and a consumer thesis — and moved directly beneath the agent setup cartridge, so the proof sits right under the call to action.

Research entities & personas Signed-out visitors can now sign in to chat with a persona right from the chat view — a clear sign-in button replaces the old dead-end notice, so it works on mobile too, and your first question is kept so you can send it the moment you're in.

Autonomous factory & operations The autonomous factory now has a richer public operations dashboard, with workflow cards, recent activity, and dedicated detail pages for each maintenance loop that make review, repair, simplification, and changelog work easier to inspect. Its header and workflow pages are cleaner, denser, and more scannable, with a canonical brand link home, a direct path back to researcher.now, and correctly rendered running-status dots and plain status pills so customers can understand factory health and maintenance loops without wading through implementation detail.

Pipeline reliability & speed Worker queue leasing now handles packless maintenance jobs without crashing, so deploy verification can keep API and background processing healthy together. Open run pages also keep hydrating through transient deploy restarts instead of getting stranded in a half-loaded skeleton state.

Site, brand & public pages Page styles now have explicit slim entrypoints instead of defaulting to the full shared bundle, so chat, docs, entities, topics, pricing, and simple article pages only load the primitives they render. Run viewer agent prompt shortcuts, the signed-out run chat composer, and the mobile run header/status/summary layout now keep their compact styling; entity, chat, topics, account, and link pages also carry the shared primitives their DOM already uses. Normal /r run pages now serve their browser document from Vercel with VM-backed metadata and export proxying, reducing the split-brain HTML/CSS deploy path. The changelog now opens on the dated release log first, with workstreams moved into a separate tab that leads with 1/3/7/30-day product-direction analysis.

Video, podcast & source analysis Trade ideas now run as a named pipeline stage with explicit lifecycle events, deterministic validation counts, and read-only tradable-universe matches across public equities, event markets, crypto venues, and token identities. A video canary also tightened extraction so broad market themes have to become specific tradable exposures or event-market queries, with obvious public-company mentions resolved even when a ticker is missing.

Payments & metered billing Awaiting-funding run pages now keep account funding in place: open the deposit dialog, check balance, and resume the report without leaving the run.

Video, podcast & source analysis Runs can now surface evidence-linked trade ideas: explicit positions and inferred market hypotheses from the corpus, with media quote and timestamp context on the run page while prices and execution stay gated behind verified resolver data.

Agent interop & the contract Shared run links are now derived from the run's own id, so the /r/ link an agent hands you always resolves to the right run — no more dead links from a fabricated or mismatched slug.

Site, brand & public pages A mistyped, stale, or fabricated /r/ run link now lands on a friendly “no cartridge in slot” page — using the Researcher skill-cartridge motif — instead of a blank loading screen. Standing topics gained a Watch targets tab: the explicit list of sources scanned in real time — X accounts, GitHub, RSS, websites, searches — with the same platform filter chips as source tables and per-platform brand icons. You can now add and remove a topic's watch targets right from the Watch targets tab, and the bootstrapping status notice moved below the topic title so the title always leads. The top nav is now consistent across the run viewer, entity pages, topics, and dashboard — same Researcher · Pricing · Docs · Topics · Account · + layout everywhere.

Report quality & evidence fidelity Quote faithfulness and topic salience are now LLM-judged gates, and embedded speaker labels are stripped from located quotes.

Pipeline reliability & speed Extraction now overlaps acquisition with speculative salience and higher concurrency — the pipeline got materially faster.

Video, podcast & source analysis Quote cards gained in-card attribution with YouTube-linked timestamps, and X search layered into acquisition with full-text articles and link expansion.

Agent interop & the contract Every terminal run is now delivered to the customer's agent by default through the research inbox.

Chat & the run workspace Extension-run proposals in chat stopped rendering as [object Object]: proposal objects are now unwrapped to their prompt text everywhere a chat message renders.

Site, brand & public pages Every page now shares the same canonical width and content edges as the run page, bringing the system, pipeline, changelog, and API docs pages in line. Source tables also gained type filters for web, video, tweet, and podcast items, with counts on entity pages, run storage, and topic watch targets so one source type no longer drowns out the rest.

Autonomous factory & operations Metrics auth moved to signed-in admin accounts, stage durations and run completion rate joined /v1/metrics, and backups stopped fighting bucket retention on latest/ pointers.

Report quality & evidence fidelity A transcript coverage gate with targeted revision, a deterministic faithfulness pass over written reports, and even transcript sampling fixed long videos losing their tails.

Payments & metered billing Cost reporting is truthful for usage-settled runs, and trial keys were removed in favor of real funded accounts.

Video, podcast & source analysis Analyze extended to X videos and podcasts on one consolidated surface; podcasts defer gracefully and duplicate in-flight requests coalesce.

Agent interop & the contract Contract versioning: every API response is stamped and agents are notified when the contract changes; fabricated share links now resolve via id-prefix fallback.

Chat & the run workspace Run-page chat docks as a right side panel on wide screens, with synced close animations and click-outside dismissal.

Site, brand & public pages Agent setup reimagined as a holographic skill cartridge, a visual deliverables section, OG images for site, runs, entities and chats, and a new tagline: turn your agent into a researcher.

Autonomous factory & operations Metrics page rebuilt with tabs, per-user spend and balances.

Payments & metered billing Analyze authorizes a $5 budget session settled to actual usage — keyed callers get a free preview and never a 402.

Video, podcast & source analysis /v1/analyze became a fully-powered run: it continues into synthesis and always creates a durable run page.

Site, brand & public pages A new brand mark — the R punched through a full ring — rolled out across favicons, and the homepage leads with Deep Research at the canonical 64rem width.

Pipeline reliability & speed Structural reliability batch: schema-derived repair, auto-retry, search self-healing, a completion contract, and a production E2E canary; runs are fitted to their budget instead of rejected on estimate.

Payments & metered billing Runs are priced at actual cost instead of tier floors, and a fund action raises a run's budget in place to keep it going.

Provider rails & failover After a provider outage took down paid calls, the LLM chain was hardened against MPP identity loss with session recovery, cooldown, retries, and real error propagation.

Video, podcast & source analysis No-speech videos complete deterministically with transcript warnings, and the Article/Video Analysis tier shipped.

Agent interop & the contract Installable skill published to npm and PyPI (Hermes plugin), with legacy-shape callers steered to the current contract.

Pipeline reliability & speed Customer run recovery fixes for extraction gates and transcript-first media runs.

Autonomous factory & operations An admin runs debug view, and the runs page went public and mobile-friendly.

Agent interop & the contract The Researcher MCP endpoint shipped and is exposed through the public domain.

Research entities & personas Entity avatars, broader profile evidence selection, durable About reports, and fixed ingestion indexing.

Autonomous factory & operations Centaur ships Linear tickets end-to-end: auto-ship by default, readable status markers, a fast lane for small tickets, and autonomous deploy recovery.

Site, brand & public pages Shared page width normalized toward one canonical shell across surfaces.

Research entities & personas Durable research entities landed in one push: entity pages, persona profiles, persona chat, X archive backfill, and About profiles that read like reports.

Autonomous factory & operations Production Postgres PITR backups with runbooks and a restore drill; Centaur gained deploy keys, reviewer guards, and verification logs.

Pipeline reliability & speed The run factory became self-healing end-to-end, with source-only extraction rescue paths for stuck runs.

Payments & metered billing A billing margin audit canary reconciles billed totals to settled channel spend, and payment-session settlement accounting was fixed.

Autonomous factory & operations An operations metrics dashboard landed.

Report quality & evidence fidelity Writer prompt consolidated from 28 instructions to 15, direct quotes render as styled quote cards, and rewritten reports are verified before completion.

Provider rails & failover Managed provider payment contract: Exa and Firecrawl run on MPP rails, with an OpenRouter API-credit fallback and Deepgram joining for audio.

Video, podcast & source analysis Deepgram transcription covers captionless videos, and failed transcript extraction is no longer fatal to a run.

Agent interop & the contract Agent Contract v1 rolled out clean-slate, with budget aliases honored.

Autonomous factory & operations Deploy stability runtime split and a documented safe production deploy workflow.

Report quality & evidence fidelity The writer now sees full ranked evidence (bestSources ordering, 500K-token budget) instead of a stripped corpus, and customer instructions are authoritative over library context.

Report quality & evidence fidelity The report writer was decomposed into a sectioned pipeline with adversarial testing, corpus merge into focused LLM passes, and every report now opens with a Signal block.

Pipeline reliability & speed Stage-based resume: continue skips re-planning and reuses corpus analysis instead of re-running the whole pipeline; extraction streams progress every 30 seconds.

Site, brand & public pages The pipeline page was redesigned stage by stage, public docs were redesigned, and this changelog page was born.

Report quality & evidence fidelity Corpus merge went hierarchical with shard retries, and rewritten reports are verified against the output contract.

Payments & metered billing Costs break down by pipeline stage and the minimum run budget dropped to one dollar.

Provider rails & failover OpenRouter MPP payment identity propagates across every LLM stage.

Payments & metered billing Customer sessions are sponsored from the treasury wallet, live billing and settlement were fixed, and run cost breakdowns shipped.

Codebase health & modularization The great split: hundreds of files broken into focused modules, with workflows and the engine promoted to the repo top level.

Pipeline reliability & speed Stop decisions are quality-gated, and budget-ended runs still complete with a report.

Video, podcast & source analysis Speaker labeling and media concept extraction became LLM-driven instead of heuristic.

Codebase health & modularization UI and runtime modularized into focused modules; the core workflow package became the engine.

Codebase health & modularization Dozens of duplicated value helpers consolidated into shared utilities.

Report quality & evidence fidelity Corpus quality overhaul — dedup, rescoring, boilerplate stripping, ranking — plus fixes for 100+ review findings across the codebase.

Provider rails & failover Content-filter refusals are detected and rerouted across Claude and GPT fallbacks with lenient JSON recovery; Exa gained find_similar, domain filters, and date ranges.

Video, podcast & source analysis Transcripts are LLM-restructured into interview format; YouTube handling is transcript-first, searching only for specific references.

Agent interop & the contract API overhaul: /research and /agent with personas and streaming, plus discovery surfaces — robots.txt, sitemap, and llms-full.txt.

Chat & the run workspace Global chat gained its sidebar everywhere, a stop button, and an iridescent focus sweep on the research input.

Video, podcast & source analysis Video reports became quote-first with timestamp links, and YouTube extraction starts straight from the home composer.

Video, podcast & source analysis From spec to shipped in a day: YouTube media runs with transcript-first reports, a transcript tab, and timestamped source references.

Provider rails & failover Exa search discovery joined Firecrawl, with costs accounted per run.

Chat & the run workspace Smoother analyst streaming, an animated sidebar, and a widened chat modal.

Agent interop & the contract An agent operating index, research contracts, and run controls for programmatic callers.

Chat & the run workspace Chat became a first-class surface: persistent start-chat sessions with shareable URLs, inline run proposals, and unified entrypoints.

Agent interop & the contract Account-linked Discord integration with customer Hermes agent routing and billing coverage.

Chat & the run workspace Run Analyst agents landed: chat with a finished run, with inline citations and durable structured answers.

Codebase health & modularization The service split into monorepo packages.

Report quality & evidence fidelity Benchmarked against OpenAI Deep Research with per-source comparisons and better eval corpus packing.

Agent interop & the contract A managed Discord research agent: direct messages, conversational chat, and persistent research memory.

Payments & metered billing Metered run billing moved to MPP sessions authorized from server wallets, with hard budget caps and full API key management.

Agent interop & the contract An agent handoff contract with funding guidance baked into 402 responses and capped planner latency.

Agent interop & the contract Share pages render for agents and run pages offer an agent handoff.

Report quality & evidence fidelity Contradiction detection landed: knowledge claims are auto-verified and contradictory facts get their own report section.

Report quality & evidence fidelity Fallback reports stopped padding themselves: boilerplate filtering, topic-specific claims, and irrelevant timelines and tension sections suppressed.

Report quality & evidence fidelity Reports learned to read as prose: plain-English sections, generic section schemas, cleaner tables, and a soft auto-reveal when a run finishes.

Pipeline reliability & speed Interrupted research jobs are requeued automatically.

Agent interop & the contract Agent setup and stored source access landed, plus an Obsidian export command.

Site, brand & public pages Homepage reframed around featured example runs and a six-step pipeline; budget-based pricing pages with a redesigned estimator.

Pipeline reliability & speed Continuation runs landed across the stack — a finished run can be extended instead of restarted.

Report quality & evidence fidelity Structured research plans replaced playbooks and the mode concept; runs honor academic-source requests.

Report quality & evidence fidelity Research loops became explicit and quality-gated, with an analyst audit pass over source extractions and library recall.

Pipeline reliability & speed Interrupted jobs recover on startup and research loops tolerate provider stalls.

Report quality & evidence fidelity First quality-first pass over the pipeline: evidence-aware quality gates, LLM candidate triage, analyst workplans, and stage-specific models.

Payments & metered billing Funded wallets are charged before any 402, and runs return watch URLs in account usage.

Provider rails & failover Provider-aware structured LLM client with OpenRouter timeouts, fallback planning models, and adaptive output budgets.

Payments & metered billing First customer payments: mppx payment challenges, account funding, and Privy key management.

Site, brand & public pages The service became Researcher, with an animated research harness diagram on the homepage.

Autonomous factory & operations v1 scaffolding: docs, CLI, and VM deploy bootstrap.