AI workforce platform — open source
An AI workforce.
You're in charge.
OpenAcme runs coworkers, not chat tabs — agents with roles, memory, and their own tools. Hand them work; they report back. On your laptop, or a server your whole team shares.
manager
Triage & routing
analyst
Research & briefs
writer
Drafts & posts
engineer
Code & CI
— dispatcher tick —
01Roster
Coworkers, not chat tabs.
Add one, retire one, change a model — you're the org chart.
---
name: Chief of staff
model: claude-opus-4
tools: [task_create, task_update, agent_list, web_search]
skills: [weekly-report]
---
You run the team's queue. When work arrives unaddressed, decide who takes it — don't relay, decide. Split anything that needs two people.
02Dispatch
Assign work and walk away.
#14
analyst
#16 weekly metrics
analyst · mon 09:00
#17 customer follow-ups
team · manager triages
#14
analyst
#15 fix flaky auth test
engineer
#14
analyst
#12 support themes
analyst
#11 release notes
writer
analyst
1Brief it like a coworker — a title and what done looks like.
2Recurring tasks re-open themselves. Dependent tasks wait their turn.
3Results come back as comments. Read them when you're back.
03Capabilities
Everything a coworker needs.
03.1Memory
What they learn, sticks
Corrections, preferences, context — saved to a Markdown file per agent, carried into every future session.
- Invoices: totals in EUR, net-30
- Weekly report ships monday 09:00
- Plain sentences, no hype
03.2Skills
Teachable in Markdown
Install capabilities from GitHub and marketplaces, or write your own.
$ openacme skills install release-notes
✓ installed — every agent can use it
03.3Models
Any model, per agent
Mix providers across the roster. Local models work too.
model:
claude-opus-4
Anthropic
03.4Teams
A manager triages
Send work to the team — the manager claims, reassigns, or splits it.
03.5Browser
A browser per agent
Own logins, own cookies. One agent's session never leaks into another's.
own profile
own logins
own profile
own logins
03.6Schedule
Work that recurs
Daily, weekly, custom — finished recurring tasks re-open themselves.
#16 weekly metrics
03.7MCP
Plugs into MCP
Every MCP server's tools, namespaced per agent or shared.
mcp-github__create_pr
mcp-slack__post_message
mcp-notion__update_page
04Console
The whole thing is also a CLI.
~/.openacme, readable without us.openacme setupconnect a provider and sign inopenacme chatterminal chat, no server neededopenacme skillsinstall, write, auditopenacme mcpwire up any MCP serveropenacme memoryinspect what an agent knowsopenacme start · stop · logsthe daemon itself05Deployment
Runs where you work.
01Your laptop
npm install, sign in, done. Everything lives in ~/.openacme — plain Markdown and SQLite you can open, read, and back up.
02A shared server
Start the daemon with --expose and your whole team works with the same workforce — one URL, secret-gated, always on.
03No platform between
There is no OpenAcme cloud. Your prompts and data go to the model provider you chose and nowhere else.
Specifications
- Runtime
- Node 18+ · macOS or Linux
- Interfaces
- Web console · CLI · headless daemon
- Models
- Anthropic · OpenAI · Google · OpenRouter · Ollama · custom
- Auth
- Claude / ChatGPT subscription, or API keys
- Storage
- Plain files + SQLite under ~/.openacme
- Network
- Loopback by default · --expose to share
- Telemetry
- None
- License
- MIT, open source
- Price
- Free — you pay only your model provider
06Start
Hire your first agent.
Your laptop or your server · your data stays yours