Open-source · MIT · v3.0.0 · coming Q3 2026

Run an AI company.
Not a prompt.

Open-source, local-first desktop app for running AI-agent organizations. Hire from 57 hand-written role specs , build a real org chart with hierarchy, file tickets, schedule work, watch the team operate live, and pull everyone into an all-hands meeting with one click.

Download v3.0.0 Browse the source
  • 57 F10 Roles
  • 10 LLM providers
  • 2,150 Tests passing
§ COCKPIT · ORG VIEW4 active · 2 idle
M
Marcus Reid
CEO · Officer
Reviewing Q3 plan
S
Sarah Chen
VP Eng · Sr Mgmt
Decomposing #214
J
Jen Park
VP Product · Sr Mgmt
Idle
D
Dev Engineer
IC · Engineering
Writing tests
R
Riley Vance
Designer · Lead
Spec for #218
T
Tyler Hu
Analyst · IC
Idle
§01 The Org

Hire from a curated F10 role library.
Not generic templates.

Six levels of hierarchy. Fifty-seven hand-written role specs across Officer, Senior Management, Management, Supervisor, Lead, and IC. Every spec is its own artifact: preferred model tier, allowed and denied tools, cooperation patterns, decision authority. Drop one into your org, set their manager, and they slot in like a person joining a team. Not like a prompt being added to a queue.

Sample role spec
--- senior-fullstack-engineer.md ---
role_id: senior-fullstack-engineer
name: Sarah Chen
level: Senior Management
preferred_model_tier: claude-opus
tools_allowed:
  - file.read
  - file.write
  - git.commit
tools_denied:
  - network.request
capabilities:
  - decompose_project
  - delegate_subtask
  - review_deliverable
---

You are a Senior Full-Stack Engineer. You write
production-grade code, mentor mid-level engineers,
and own delivery of features end-to-end…
Six hierarchy levels
Officer CEO · COO · CTO · CFO · CMO
Senior Mgmt VP Engineering · VP Product · VP Marketing
Management Eng Manager · Product Manager · Design Director
Supervisor Tech Lead · Design Lead · QA Lead
Lead Senior Engineer · Senior Designer · Senior Researcher
IC Engineer · Designer · Analyst · Writer · QA
§02 The Cockpit

Watch the team operate. Live.

Five dashboard subviews (Cards, Timeline, Stream, Floor, Org) show the company in motion. Token streams from every active turn, kanban tickets sliding from Open to In Progress as the orchestrator dispatches them, telemetry per provider and per model with date-range filters, costs computed at call-site. One Cmd+K opens the natural-language palette: fourteen structured intents, destructive-action confirmation gate, agentic-loop fallback for free-form questions.

Open03
TKT-218
Wire Cmd+Shift+K to copilot ask flow
RV · P2
TKT-219
Add tabular-nums to telemetry totals
unassigned · P3
TKT-221
Provider router cost rounding for Groq
unassigned · P2
In Progress02
TKT-214
Decompose Q3 launch project into milestones
SC · P1
TKT-216
Migrate FTS5 index to canonical name
DE · P2
Blocked01
TKT-211
Pinned Ollama version on macOS arm64
DE · waiting on upstream
Done03
TKT-208
Copilot weights changed audit event
SC · closed
TKT-205
Ticker pause-aware on meeting.started
DE · closed
TKT-204
RAG attribution dedupe on SHA256
DE · closed
You don't manage prompts. You run a company. - Team-X README · v3.0.0
§03 Intelligence Layer

Ask anything. Get a grounded answer.

RAG-grounded turns. ReAct-style agentic loop with hard step, token, and wall-clock budgets. A proactive Copilot that runs every five minutes asking "what's wrong with this company right now?" and surfaces insights across operational, cost, org, workflow, and anomaly categories.

§04 Privacy Posture

Local-first by default. Zero phone-home.

Your data lives in a SQLite database and a filesystem vault on your machine. Secrets sit in the OS keychain via keytar, never in plaintext config files. Choose which provider tiers your agents are allowed to use. Run the entire org on Ollama with no cloud dependency at all. Updates are user-triggered. No analytics. No telemetry. No crash reports. No exceptions.

Ollama
Local · privacy-pure
llama3.1:8b · phi3 · mixtral · qwen2.5
LM Studio
Local · privacy-pure
any GGUF · MLX
Groq
Open-source cloud
llama-3.1-70b · mixtral-8x7b
OpenRouter
Open-source cloud
mixtral · llama-3 · deepseek
Together
Open-source cloud
llama-3.1-405b · qwen2.5-72b
Fireworks
Open-source cloud
deepseek · qwen · llama-3
OpenAI-compat
Bring your own
any compatible endpoint
Anthropic
Proprietary cloud
claude-opus · claude-sonnet · claude-haiku
OpenAI
Proprietary cloud
gpt-5 · gpt-4o · gpt-4o-mini
Google
Proprietary cloud
gemini-2.5-pro · gemini-flash
§05 Install

Three platforms. No account.

Native installers for Windows, macOS, and Linux. Or build from source with pnpm. No sign-up wall. No credit card on file. Run on Ollama with zero cloud, full feature parity, fully offline.

Windows
Team-X-Setup.exe
x64 · arm64 · 84 MB
Download
macOS
Team-X.dmg
Intel · Apple Silicon · 92 MB
Download
Linux
Team-X.AppImage
x64 · .deb available · 78 MB
Download
Or run locally via Ollama
# Prereqs: Node 20+, pnpm 9+
git clone https://github.com/git-rocky-stack/team-x.git
cd team-x
pnpm install
pnpm dev

# Local model
ollama pull llama3.1:8b
ollama serve
§06 Field Notes

Build notes from running an AI company.

A weekly note on what we're learning shipping Team-X: the architectural choices, the role-spec patterns that work, the failure modes we hit, and the operational metrics from running our own AI org. No marketing. No fluff. Cancel any time. RSS first-class.

Run an AI company. Not a prompt.
Download v3.0.0 Read the docs Star on GitHub