Open Coworkers · Free · MIT · Open sourceA researcher and a writer that hand work to each other.

Specialist AI coworkers with their own permissions, a shared workspace with versions, and explicit hand-offs.

Get a CompanyFabric keyGitHub repo coming soon
With no key it runs a clearly labelled demo.
Open Coworkers running its built-in demo on sample data
Sample dataOpen Coworkers, running its built-in demo.

How it works

From a request to finished work

  1. R To Ria, Researcher

    Research the Model Context Protocol, save notes, then have Wes write a short newsletter summary.

    Ask a follow-up…

    01

    Give Ria a task

    Ask the Researcher in plain words, and say who writes it up.

  2. R Ria saved to the workspace

    Research notes (DEMO)

    • MCP standardizes how apps give tools to models 1.
    • Language models are trained on large text corpora 2.

    RW Handed off to Wes, Writer

    Write a three-sentence summary of the research notes for a newsletter, keeping the citations.

    02

    She researches and hands off

    Ria saves cited notes to the workspace and passes the write-up to Wes.

  3. Saved to the workspace
    Title
    Newsletter summary (DEMO)
    By
    Wes, Writer

    MCP gives models a standard way to use tools 1. The models themselves are trained on large amounts of text 2. Together they make agents that can act, not just answer.

    03

    Wes writes, you approve

    Wes works only from the workspace, and his save waits for you.

Sample data Scenes use the starter's built-in demo data.

Approvals

Each coworker has its own permissions.

Ria works on her own. Every document Wes saves waits for your approval, and he cannot search the web: his policy denies it on the server, not a line in his prompt.

Wes, Writer

  • Read documentdoc.readAllowed
  • Save documentdoc.writeAsks you
  • Hand offteam.handoffAllowed
  • Web searchsearch.webDenied
Wes wants to save a document
Title
Newsletter summary (DEMO)

MCP gives models a standard way to use tools 1. The models themselves are trained on large amounts of text 2. Together they make agents that can act, not just answer.

Sample data The app's approval card.

Screenshots

The real app

A finished demo turn in Open Coworkers, on sample data
Sample dataA finished demo turn.
Open Coworkers's side panel (Workspace, with the summary Wes saved), on sample data
Sample dataThe side panel: Workspace, with the summary Wes saved.

Use cases

What you can hand it

Research to brief

Ria gathers the facts, Wes writes the page.

One ask to start with

“Research what changed in the EU AI Act for 2026, save notes, then have Wes write a one-page brief.”

Source gathering

Credible sources, saved where the team can use them.

One ask to start with

“Find three credible sources on the four-day work week and save them to the workspace.”

Post from notes

Wes turns saved research into a post.

One ask to start with

“Draft a short LinkedIn post from the latest research notes in the workspace.”

Fill the gaps

Wes hands missing facts back to Ria instead of guessing.

One ask to start with

“Write a one-paragraph update from the research notes, and hand any missing facts back to Ria.”

Compared

Start from working code

Open Coworkers

  • PermissionsA policy per coworker per tool, enforced on the server.
  • Hand-offsExplicit hand-offs; the next turn starts when the current one ends.
  • WorkspaceShared documents, with old versions kept.
  • ApprovalsOne tool layer asks you before every gated action, enforced on the server.
  • Audit logAn audit row is written before every tool call.
  • RuntimesLite on Vercel or Hermes in Docker, from the same app.
  • ModelsAny OpenAI-compatible endpoint with tool calling. CompanyFabric by default.

Building it yourself

  • PermissionsPrompts that ask an agent not to use a tool.
  • Hand-offsA queue and shared state between agents, built yourself.
  • WorkspaceStorage and versioning designed from scratch.
  • ApprovalsApproval flows built into each tool, kept in step with the prompt.
  • Audit logLogging added around every tool by hand.
  • RuntimesOne agent loop; a second runtime is a second integration.
  • ModelsProvider wiring and switching are yours to build.

FAQ

Questions

Lite or Hermes: which should I run?

Both run from the same app. Lite runs the agent loop in the Next.js server (pnpm dev), and Vercel works. Hermes runs each turn in a Hermes Agent container (v0.22.5, pinned by digest) (docker compose up) and calls the app's tools back over MCP, so policies, approvals and the audit log are the same. Hermes adds memory across threads, and needs a long-running container.

Which model key do I need?

Any OpenAI-compatible chat.completions endpoint with tool calling. The default is CompanyFabric: one key for every major model, billed per token from a prepaid balance, and companyfabric/auto picks a suitable model per request. OpenAI directly, or a local server such as Ollama, LM Studio or vLLM, works too.

OPENAI_BASE_URL=https://api.companyfabric.com/v1
OPENAI_API_KEY=sk-cf-…
MODEL=companyfabric/auto
Can I try it without any keys?

Yes. With no model key it runs a DEMO: scripted turns, clearly labelled, with no model or search call. The workspace tools, permissions, approvals and hand-offs in it are the real ones, so Ria's notes land in the workspace and Wes's draft waits for your approval.

Can I self-host it?

Yes. It is a Next.js app that keeps its data in one SQLite file: run it on your own machine or server, on Vercel (lite), or with Docker Compose (Hermes). There is no login: the compose file binds the app to 127.0.0.1, so put it behind your own authentication before exposing it.

Can I add coworkers?

Yes: add an entry to coworkers.config.ts with a name, role, instructions and a policy per tool. Names, roles and instructions can also be edited in the app's Team tab; tool policies only in code.

Is it ready for production?

It is early. CI checks every change: Install, typecheck, unit tests and production build; Hermes starts with the app's tools loaded (no model is called in CI); Contract tests: permissions per coworker, approvals, workspace versions, hand-offs. Implemented but not yet run live: A live Ria to Wes hand-off with a real model; Search through Parallel or Exa; A Hermes turn with a real model.

What is the licence?

MIT, © 2026 Creative Tim: read it, change it and ship it. The UI is adapted from AI Elements (Apache-2.0) and shadcn/ui (MIT). Not affiliated with any company whose products offer AI coworkers.

Technical details

Lite and Hermes, capability by capability
CapabilityLiteHermes
Specialist coworkers with their own tool permissionsYesYes
Shared workspace documents, with versionsYesYes
Hand-offs between coworkersQueued: the next coworker starts when the current one finishesYesYes
Approval before side effectsYesYes
Audit row before every tool callYesYes
Web search with numbered citationsYesYes
Memory across threadsHermes memory is shared by the whole team (one Hermes profile)NoYes
Files, terminal, browser in a sandboxEnable toolsets in hermes/config.yamlNoOff by default
Recurring responsibilities (cron)Hermes cron works; this UI does not manage schedules yetNoNot in the UI yet
Runs on serverless hosting (Vercel)YesNo
Environment variables
VariableDefaultWhat it is
OPENAI_BASE_URLhttps://api.companyfabric.com/v1Any OpenAI-compatible chat.completions endpoint with tool calling.
OPENAI_API_KEYnoneThe model key. With none set, the app runs a labelled DEMO.
MODELcompanyfabric/autoThe model id. companyfabric/auto picks a suitable model per request.
SEARCH_PROVIDERparallelparallel (free, rate-limited, no key) or exa. companyfabric is not released yet.
PARALLEL_API_KEYnoneLifts Parallel's free-tier limits.
EXA_API_KEYnoneNeeded when SEARCH_PROVIDER=exa.
COMPANYFABRIC_SEARCH_ENABLEDfalseCompanyFabric search is not released yet; leave it off.
RUNTIMEnonelite or hermes. Empty: hermes when HERMES_URL is set, lite otherwise.
HERMES_URLnoneSet for you by docker-compose.yml.
HERMES_API_KEYnoneHermes only. Generate with: openssl rand -hex 32
TOOL_TOKENnoneHermes only: lets Hermes call the app's tools at /api/mcp. Generate with: openssl rand -hex 32
DATA_DIR./dataWhere the SQLite database lives.
What is proven

Early. Checked in CI on every change: Install, typecheck, unit tests and production build; Hermes starts with the app's tools loaded (no model is called in CI); Contract tests: permissions per coworker, approvals, workspace versions, hand-offs.

  • implemented-untestedA live Ria to Wes hand-off with a real model
  • implemented-untestedSearch through Parallel or Exa
  • implemented-untestedA Hermes turn with a real model
  • not yetManaging recurring responsibilities from the UI

Run Open Coworkers on one key

Free and MIT. Add a CompanyFabric key for live work: companyfabric/auto picks a suitable model per request, and new accounts get starter credit.

Get a CompanyFabric keyGitHub repo coming soon

MIT

License

Open Coworkers is MIT-licensed, © 2026 Creative Tim. Its UI is adapted from AI Elements (Apache-2.0) and shadcn/ui (MIT).

Not affiliated with any company whose products offer AI coworkers.

All templates →