Product
One Control Room for AI Teams
Foxwel OS is the self-hosted coordination and accountability layer for teams running AI coding agents across client projects. It launches the agents, watches their worktrees and keeps the record.
The workflow
Launch. Coordinate. Reuse Context. Record.
- 01
Launch
Start Claude Code, Cursor, Antigravity or a local model against a scoped task, in its own Git worktree, one agent per service.
- 02
Coordinate
Foxwel OS reads live Git state across every active worktree and flags files being changed in more than one place, by repository and revision, while it is still cheap to reassign.
- 03
Reuse Context
A cross-repository knowledge graph, durable memory and portable skills, so a new session starts from written context instead of re-exploring the codebase.
- 04
Record
Every session logged: who launched it, which agent and provider, which repo and branch, and usage figures where the agent or vendor reports them.
Coordinate
See Overlaps Before the Merge
Every agent works on its own branch. Foxwel OS shows which agent is editing which file, in which repo, right now, and flags the same file changing in two places.
src/auth/guard.ts
Cursor · fix-login-redirect
src/checkout/flow.ts
Claude Code · checkout-refactor
api/routes/orders.ts
Codex CLI · api-pagination
busy held by Cursor (fix-login-redirect)
Isolated Worktrees
Every agent works in its own Git worktree on its own branch. Parallel work without stepping on each other.
Live Activity
See which agent is editing which file, in which repo, right now.
Overlap Warnings
Same file changed in two active worktrees? You see it in minutes, keyed on repo and revision, not at merge time.
Merge Ordering
A suggested integration order, based on how much each branch overlaps the others.
Reuse context
Hand Off, Don't Start Over
Pass a task with its plan, files and notes to any agent. Plan on Claude Code, finish on Antigravity, nothing restarts.
Plan and think
Claude Code
foxwel pass my-task --to cursor
HANDOFF.md
- Objective
- Plan
- Checklist
- Touched files
- Git state
- Graph excerpt
Plus the notes the next agent needs
Edit and finish
Cursor
foxwel take my-task▍
Handoffs That Carry Context
The plan, the files and the notes travel with the task to the next agent.
Memory and Notes
Durable facts and handoffs tied to the code, shared across agents and sessions.
Cross-Repo Knowledge Graph
One index across all your services. Rebuilt in seconds, exportable between machines.
Portable Skills
A curated skills registry, installed once, working across Claude Code, Cursor and Antigravity runners.
Control and record
Agents Launch Agents. You Approve.
Agents can launch other agents. Every launch waits for a human yes, and every session lands in the ledger.
Claude Code · ship-search-v2
RunningCodex CLI · write-search-tests
✓DoneAider · update-search-docs
RunningLaunch request
Gemini CLI · index-perf-pass
Agent Approvals
Agents can launch other agents. Every launch waits for a human yes.
Session Ledger
A record of every observed run: who, which agent, which provider, which repo and branch, how long, and usage where available.
Client Attribution and Reports
Map repos and sessions to client engagements and export a per-client summary in minutes.
Model Routing and Local Models
Write routing as policy: routine work to local or small models via Ollama or LM Studio, hard reasoning to frontier models.
Launch Policies, Budgets and Roles
Allow-lists and budget checks applied when a session starts. Roles and permissions for multi-user teams.
Client-Scoped Boundaries
Index, memory, handoffs, skills and exports scoped per client, so one engagement can't retrieve another's context.
Product tour
See Every Feature at Work
The real app sidebar, working. Check a busy file, approve a launch, review a merge.
- Active sessions
- 4
- Tasks
- 11
- Conflicts
- 2
- Claude Code
api-server · s-41
Planning the orders split
- Cursor
admin-panel · s-42
Running admin tests
- Antigravity
storefront · s-43
Writing component tests
- Ollama
driver-app · s-44
Renaming props
4 sessions · 4 worktrees · updates as agents work
Every active session, agent, task and repo on one screen, with live counts.
Learn moreLive demo
Try It Right Here
Click a real foxwel command to see what it does. Everything is also in the desktop app.
$foxwel pass checkout-refactor --to cursor▍
✓ Packed session into HANDOFF.md
objective · plan · checklist · touched files
git state · code-map excerpt · notes
→ next: foxwel take checkout-refactor
Sample output · install to run on your repo
Privacy and architecture
It Starts the Agents. It Is Not on the Wire.
Foxwel OS runs on your own machine or office server. Your hardware, your keys. It launches agents, briefs them and watches their worktrees. Agents call their model providers directly.
Your Hardware
Self-hosted on a developer machine or a shared server on your network.
Your Keys
Agents use your own model contracts and keys, and call providers directly.
No Hosted Copy
No Foxwel-hosted copy of your source. We broker no inference and take no margin on it.
Not a Replacement
Not a model, not an editor, not a replacement agent. Developers keep their IDE and their tools.
No Telemetry
The app sends no usage analytics about your code or your sessions to us.
A Session Record Shows
- Who ran it and when
- Which agent and provider
- Which repo, branch and files
- Usage figures where reported
It Does Not Claim to Show
- The exact content sent to a provider
- Activity outside supported agents: raw terminal runs, IDE autocomplete, copy-paste into chat
- Your vendor invoice
- Supported agents
- 8
- Repos indexed on our pilot estate
- 6
- Nodes in the pilot knowledge graph
- 24,541
- Lines of your source on Foxwel servers
- 0
