<- all tokdocs
2026.10.23@buildwithneej

OpenRig's seven-seat factory team is exactly what the video says it is, but the repo is mvschwarz/openrig at 8,168 stars, not the 5,000 the narration claims, and it boots four agent runtimes rather than two

AI
Watch on TikTok

View on TikTok ->

A 37.150 s vertical clip at 1080x1920 in a mov/mp4 container, HEVC video at 828,478 bps and 30/1 fps across 1,114 frames alongside AAC stereo at 44,100 Hz and 64,345 bps across 804 frames, 4,176,667 bytes total at an overall 899,416 bps, carrying TikTok's bytevc1_1080p_899392-1 format id, posted at epoch 1792769046 which is 2026-10-23T15:24:06Z by @buildwithneej (nickname "Neej") over an "original sound" credit attributed to Neej, with an 855-character 155-word description and a 72-character title, standing at 6,309 views, 272 likes, 18 comments and 74 reposts when metadata.json was written at 2026-10-24T23:10:23Z. I read 17 of the 19 extracted frames (001, 002, 003, 005, 006, 008, 009, 010, 011, 012, 013, 014, 015, 016, 017, 018 and 019, spaced 2 s apart from roughly 0 s to 36 s) against a 13-segment, 152-word transcript. Frame 001 at 0 s overlays the white hook card "Turn Claude + Codex into a 7-agent dev team" on a screen recording of a terminal UI whose table lists pods dev, orch, review, dev50 and review50 with seats carrying runtime codes cl and cx, models fable-5-1 and gpt-6-astra, context percentages and states, footed by "17 seats · 6 working · 1 need attention", above a rendered README showing two install lines, "# Claude Code / npx skills add https://github.com/mvschwarz/openrig/tree/..." and "# Codex / npx skills add https://github.com/mvschwarz/openrig/tree/...". Frame 002 at 2 s shows the same table with the MODEL column reading "Opus 5.5" and "gpt-6-a..." and a footer line "from the repo · example data". Frame 003 at 4 s pairs a gold star graphic and the burned-in figure "5,000" with the topology graph behind it. Frame 005 at 8 s is a designed card reading "RUNNING AI AGENTS · ONE VS A TEAM / one agent: easy. / 1 AGENT · 1 LINE" over "a team: the hard part. / 5 AGENTS · EVERY LINE BETWEEN THEM", disclaimed in small caps as "ILLUSTRATION OF THE IDEA, NOT A MEASUREMENT". Frame 009 at 16 s shows the TUI header "herdr spaces | OpenRig TUI" with an explorer tree rooted at "macbook-pro" and a child node "factory". Frame 010 at 18 s shows the rendered README section "Teams" with the sentence "Three teams grow with you, and the kernel operator helps you pick one:" and the bullets starter, workshop and factory, while a caption overlay prints "starter", "pm" and "factory". Frame 011 at 20 s is a card reading "FACTORY TEAM · 7 SEATS / Seven agents, three pods." Frames 012 and 013 at 22 s and 24 s show a lower-third card "TEAM · FACTORY / the factory team: seven agents" naming seven seats in three columns, Orchestration (lead, advisor), Development (build, qa, design) and Review (r1, r2), each tagged codex or claude-code. Frame 014 at 26 s shows the README paragraph beginning "Specialist teams: code-review (two independent reviews), research (an analyst and a synthesizer) and pm". Frames 015 and 016 at 28 s and 30 s show "TEAM · STARTER / build + review" with "CLAUDE CODE build dev-build@starter" joined by a "delegates_to" arrow to "CODEX review dev-review@starter", subtitled "Claude builds · Codex reviews". Frame 017 at 32 s shows two circles labelled "Claude Code" and "Codex" over a README line reading "Claude Code, Codex and Pi work together as a persistent team", and frame 019 at 36 s closes on "COMMENT TEAM down below and I'll send you the repo" over a TikTok comment box. The spoken argument is that running a single AI coding agent is easy and coordinating several is hard, that OpenRig solves this by booting each agent into a Claude or Codex session, assigning it a job and letting the agents message each other under a live dashboard, and that you choose between a seven-agent factory team, a PM team and a two-agent starter team. The transcript's last segment runs 36.86 s to 37.08 s against a 37.150 s container duration, so nothing is timestamped past the end of the audio and no segment is a Whisper hallucination, though segment four at 10.78 s renders the product name as "open rig" in lower case before segment eleven at 33.64 s gets it right as "OpenRig".

The seven-agent factory claim is correct, and the shipped YAML names every seat

At 19.82 s the video says "The factory team has seven agents, a lead, advisor, a builder, QA, a designer, and two reviewers." That survives a line-by-line check. The file packages/daemon/specs/rigs/launch/factory/rig.yaml on the main branch declares three pods. The orch pod, labelled Orchestration, holds lead (agent_ref: local:../../../agents/orchestration/orchestrator, runtime: codex) and advisor (agent_ref: local:../../../agents/product-management/pm, runtime: claude-code, profile: advisor). The dev pod, labelled Development, holds build (local:../../../agents/development/implementer, runtime: claude-code), qa (local:../../../agents/development/qa, runtime: codex) and design (local:../../../agents/design/product-designer, runtime: claude-code). The review pod holds r1 and r2, both pointing at local:../../../agents/review/independent-reviewer, with r1 on claude-code and r2 on codex. Two plus three plus two is seven members, and the lower-third card in frames 012 and 013 reproduces that exact grouping with the same runtime labels.

The repo's own summary field in that file reads "A product factory with a lead, a PM advisor, build, QA, design, and two independent reviewers. Bring ideas to the advisor; the lead runs the team." The one detail the narration flattens is that the advisor is not a generic advisor seat. It resolves to the product-management agent spec running under an advisor profile, which is why the README calls it a PM advisor. The video's role list is otherwise an accurate reading of the file.

"Over 5,000 stars" was stale before the video was uploaded; GitHub reported 8,168, and the clip landed 72 minutes after the 0.7.0 release

At 4.24 s the narration says "This repo has over 5,000 stars," and frame 003 at 4 s burns the figure "5,000" into the picture next to a star graphic. The GitHub REST API response for https://api.github.com/repos/mvschwarz/openrig, captured at 2026-10-24T23:13:55Z, returned stargazers_count of 8,168, along with 666 forks, 29 subscribers and 147 open issues. The video's own TikTok description, written by the same creator, says "over 8,000 people have starred it on GitHub," so the audio track contradicts the caption that ships with it. Treat the spoken 5,000 as a number recorded earlier and never re-cut, and treat 8,168 as the figure at the capture time above. A star count moves, so the figure is only meaningful with its timestamp attached.

The repo's oldest commit, f3a97b0277, is dated 2026-03-23T07:09:13Z by Mike Schwarz and is labelled an initial project scaffold. The GitHub repository record carries a created_at of 2026-04-01T06:22:40Z, so roughly nine days of history preceded the repo going public. The npm package @openrig/cli was first published as 0.0.1 on 2026-04-06T21:42:11Z and has 65 versions on the registry. The GitHub releases endpoint returns 50 releases, the oldest listed being v0.1.12 on 2026-04-19T20:04:59Z and the newest v0.7.0, published at 2026-10-23T14:12:22Z, three minutes after npm received 0.7.0 at 2026-10-23T14:09:47Z. The TikTok posted at 2026-10-23T15:24:06Z, 71 minutes and 44 seconds after the GitHub release went live. The video never claims to cover the release, and the proximity explains why the repo's state in the footage is current while the star figure in the audio is not.

The licence the video alludes to with "It's open source" in its description resolves to Apache-2.0 on both GitHub and the npm registry entry for @openrig/cli@0.7.0. The platform requirements the description lists are stated in the README: "You need Node.js 22 or 24 (22 on Apple silicon), tmux, macOS or Linux (WSL2 on Windows), and Claude Code or Codex signed in. One provider is enough to start."

"A Claude or Codex session" undercounts the runtimes OpenRig actually boots

At 10.78 s the video says the repo "boots each agent into a Claude or Codex session," and frame 017 at 32 s puts two circles on screen, "Claude Code" and "Codex." The README section "How It Works" lists four under Runtimes: "Native Claude Code and Codex sessions, terminal nodes, and Pi and Oh My Pi via RPC runners." The repo's GitHub description is more explicit still: "Build your own network of agents from Claude Code, Codex and Pi: persistent teams with roles, shared context and owned work." The rig specification at docs/reference/rig-spec.md documents each one with its own heading, including "### Pi (runtime: pi)", which states that "runtime: pi launches the Pi coding agent (pi --mode rpc) through OpenRig's RPC runner," and "### Oh My Pi (runtime: omp)", which states that Oh My Pi "is separate from runtime: pi; OMP does not use Pi's --name or --approve flags." The same document defines runtime: terminal for infrastructure processes such as servers, log tails and build watchers, and notes those are "not agent runtimes."

The video's own screen recording gives this away. The README text visible behind the circles in frame 017 reads "Claude Code, Codex and Pi work together as a persistent team," with Pi sitting right there on screen while the narration names only two. The claim is not wrong about what Claude Code and Codex can do inside a rig. It is incomplete about what the project supports.

The three-team list on screen says workshop; the narration substitutes pm

At 25.4 s the video says "The PM team builds mock-ups and prototypes," placing it between the factory team and the starter team as if the three were peers. Frame 010 at 18 s shows the README section the video is scrolling, and that section reads "Three teams grow with you, and the kernel operator helps you pick one:" followed by starter ("a builder and a reviewer for one bounded change"), workshop ("a lead, a builder, QA and a reviewer for ongoing work in one repository. It's a rig bundle, installed from its pinned listing") and factory ("seven agents (a lead, an advisor, build, QA, design and two independent reviewers) for sustained product work"). The caption overlay in that same frame prints "starter", "pm" and "factory", replacing workshop with pm.

Eight seconds later the video scrolls far enough to contradict itself. Frame 014 at 26 s shows the next README paragraph: "Specialist teams: code-review (two independent reviews), research (an analyst and a synthesizer) and pm (a product lead, a researcher and a builder for prototypes)." The pm rig is a specialist team, and it lives at packages/daemon/specs/rigs/focused/pm/rig.yaml, under focused/, while starter and factory live under launch/. Its summary reads "Product management rig with a PM lead running the 8-step feature flow (validate, context, requirements, mockups, review, exec summary, handoff). Includes a researcher for competitive/regulatory/customer intelligence and a coder for UI mockups and prototypes." Three members are declared: pm.lead on claude-code, pm.researcher on codex and dev.build on claude-code.

So the mockups-and-prototypes description is accurate for one of the pm rig's three seats, and the framing that puts pm in the shipped three-team ladder is wrong. The team the video skipped, workshop, is the one the README singles out for broad permissions, noting that it "ships with broad access."

The starter team's wiring matches the video, and the "accounts you already have" line holds only for that team

At 27.58 s the video says "a starter team has Claude build while Codex reviews every change," and frames 015 and 016 draw it as dev-build@starter under a CLAUDE CODE header pointing through a delegates_to edge to dev-review@starter under a CODEX header. The file packages/daemon/specs/rigs/launch/starter/rig.yaml declares a single dev pod labelled Project with build (local:../../../agents/development/implementer, runtime: claude-code) and review (local:../../../agents/development/qa, runtime: codex, model: gpt-6-astra), joined by - kind: delegates_to / from: build / to: review. The direction, the runtimes and the seat addresses all match the card on screen.

Two details the video compresses are worth naming. The reviewer seat resolves to the QA agent spec at agents/development/qa, not to the agents/review/independent-reviewer spec the factory team uses for r1 and r2, so the starter reviewer is a QA agent wearing a review seat. And the starter spec pins a model string, gpt-6-astra, for that seat, which is why that same string appears in the MODEL column of the TUI footage in frames 001 and 008.

At 31.22 s the video says "This runs on the Claude and Codex accounts that you already have." The README supports the weaker version of that: "one working Claude Code or Codex account: one is enough, since the kernel operator adapts the shipped starter team to whichever you have." The factory team is a different case. Its rig.yaml hard-codes runtime: codex for the lead, runtime: claude-code for the advisor, and a mix across the dev and review pods, so running the seven-seat team exactly as shipped needs both providers signed in. A single-provider user gets the adapted starter team, not the factory. The repo also ships rig-claude-only.yaml and rig-codex-only.yaml variants for the kernel rig at packages/daemon/specs/rigs/launch/kernel/, which is the mechanism behind that adaptation.

The dashboard is a terminal UI, and the footage is the repo's own fictional demo data

At 16.48 s the video says "You can watch your whole team live in a dashboard." OpenRig does ship one. The README describes the system as "a local daemon + CLI + terminal UI + MCP server, built on tmux," and lists the TUI's surfaces as "Topology explorer, table and graph views, seat details, Specs, Projects, Terminals, Feed, and System." Frame 009 at 16 s shows that chrome directly, with the header "herdr spaces | OpenRig TUI" and the tab row "[ TABLE ] RECENT OVERVIEW GRAPH HEALTH." The word "dashboard" is doing some work here, because the README adds that "only the older React web UI is in maintenance mode, with best-effort support," so the maintained surface is the terminal UI rather than a browser dashboard.

The footage behind the narration is not a run of the creator's own agents. Frame 002 at 2 s carries a small footer reading "from the repo · example data," and the README's own screenshot caption says "Captured from the interactive TUI demo using fictional project data." The README does include one piece of genuine footage, a GIF whose alt text reads "The OpenRig TUI: the build rig as a graph, then as a table of seats with runtime, model, context and state, then one seat in detail (real recording, 10 seconds)." No terminal command executing, no install run and no agent output appears in any frame I read. The video shows the README and the demo data, and it labels the demo data on screen.

One detail in that demo data checks out against live documentation. The MODEL column in frames 001, 002 and 008 shows fable-5-1 and Opus 5.5. Anthropic's models overview at platform.claude.com lists Claude Fable 5.1 with the API id claude-fable-5-1 and Claude Opus 5.5 with the API id claude-opus-5-5 among the current lineup, alongside Claude Sonnet 5.5 and Claude Haiku 5.5. The project data in the TUI is fictional and the model identifiers printed next to it are real.

Key Takeaways

  • Verified: The factory team really is seven seats. packages/daemon/specs/rigs/launch/factory/rig.yaml declares orch.lead, orch.advisor, dev.build, dev.qa, dev.design, review.r1 and review.r2, matching the lead, advisor, builder, QA, designer and two reviewers the video names at 19.82 s.
  • Verified: The starter team wires a Claude Code builder to a Codex reviewer through a delegates_to edge, exactly as the video says at 27.58 s and as frames 015 and 016 draw it.
  • Correction: The video says "over 5,000 stars" at 4.24 s and burns "5,000" on screen. The GitHub API reported 8,168 stars for mvschwarz/openrig at 2026-10-24T23:13:55Z, and the video's own TikTok description says "over 8,000."
  • Correction: The video says the repo boots each agent "into a Claude or Codex session" at 10.78 s. The README lists four runtimes, Claude Code, Codex, Pi and Oh My Pi, plus terminal nodes, and docs/reference/rig-spec.md documents runtime: pi and runtime: omp with their own launch behaviour.
  • Correction: The video presents "the PM team" at 25.4 s as one of the choosable teams. The README section visible in frame 010 lists starter, workshop and factory as the three teams, and frame 014 shows pm listed separately under "Specialist teams" alongside code-review and research.
  • Partial correction: The claim at 31.22 s that OpenRig "runs on the Claude and Codex accounts that you already have" is true for the starter team, which the kernel operator adapts to whichever single provider is signed in. The factory team pins both claude-code and codex seats in its YAML and needs both.
  • Clarification: The TUI footage is the repo's own demo. Frame 002 carries the footer "from the repo · example data," and the README screenshot caption reads "Captured from the interactive TUI demo using fictional project data." No install, command execution or live agent output appears in any frame I read.
  • Clarification: The maintained visual surface is a terminal UI built on tmux. The README states that "only the older React web UI is in maintenance mode, with best-effort support," which qualifies the word "dashboard" used at 16.48 s.
  • Context: The model strings in the demo table are real. Anthropic's models overview lists Claude Fable 5.1 (claude-fable-5-1) and Claude Opus 5.5 (claude-opus-5-5) in the current lineup, matching the fable-5-1 and Opus 5.5 labels in frames 001, 002 and 008.
  • Context: The repo is Apache-2.0, owned by Mike Schwarz (mvschwarz), first committed on 2026-03-23, made public on 2026-04-01, and released as v0.7.0 on 2026-10-23 at 14:12:22Z. The TikTok posted 72 minutes later at 15:24:06Z.

Resources

Published October 23, 2026. Writeup generated from a favorited TikTok.