Agentic Development

FossilRepo is built for AI-assisted development at scale. Traditional Git forges impose rate limits that cripple agent workflows. FossilRepo eliminates these bottlenecks.

The Problem

AI coding agents make dozens of API calls per task. On GitHub: - 5,000 API calls/hour limit (agents burn through this in minutes) - 30 search requests/minute - Webhook delivery delays - Actions queue congestion

Multiple agents working in parallel hit rate limits within seconds.

The Solution

AI Agent (Claude Code, Cursor, etc.)
  |
  v
FossilRepo MCP Server / API    <-- zero rate limits
  |
  v
Fossil repos (.fossil SQLite)  <-- local disk, instant
  |
  v (scheduled, batched)
Git Mirror --> GitHub           <-- rate-limit-aware sync

Agents work against FossilRepo locally with no limits. Changes sync to GitHub on a schedule. GitHub becomes a downstream mirror, not the bottleneck.

Connecting AI Tools

MCP Server (Recommended)

The MCP server gives AI tools native access to all FossilRepo capabilities.

pip install fossilrepo
fossilrepo-mcp

Claude Code config:

{
  "mcpServers": {
    "fossilrepo": {
      "command": "fossilrepo-mcp"
    }
  }
}

17 tools available: browse code, read files, search, manage tickets, view timeline/diffs/blame, create tickets, run SQL queries.

JSON API

For tools without MCP support:

curl -H "Authorization: Bearer frp_abc123..." \
  http://localhost:8000/projects/myproject/fossil/api/timeline

Batch API

Reduce round-trips by 25x:

POST /api/batch
{"requests": [
  {"method": "GET", "path": "/api/timeline"},
  {"method": "GET", "path": "/api/tickets", "params": {"status": "Open"}},
  {"method": "GET", "path": "/api/wiki/Home"}
]}

The Agent Workflow

1. Discover Work

Browse open, unclaimed tickets:

GET /api/tickets/unclaimed

2. Claim a Ticket

Atomic claiming prevents two agents from working on the same thing:

POST /api/tickets/<uuid>/claim
{"agent_id": "claude-session-abc"}

Returns 200 if claimed, 409 if already taken by another agent.

3. Create an Isolated Workspace

Each agent gets its own Fossil branch and checkout directory:

POST /api/workspaces/create
{"name": "fix-auth-bug", "agent_id": "claude-session-abc"}

No interference with other agents or the main branch.

4. Do the Work

Read code, understand context, make changes, commit. All via MCP tools or API:

POST /api/workspaces/fix-auth-bug/commit
{"message": "Fix null check in auth middleware"}

5. Submit for Review

POST /api/reviews/create
{
  "title": "Fix null pointer in auth module",
  "description": "The auth check was failing when...",
  "diff": "--- a/src/auth.py\n+++ b/src/auth.py\n...",
  "workspace": "fix-auth-bug"
}

6. Review and Merge

Another agent (or human) reviews:

POST /api/reviews/<id>/approve
POST /api/reviews/<id>/merge

7. Release the Claim

POST /api/tickets/<uuid>/submit
{"summary": "Fixed by closing the null check gap"}

Real-Time Coordination

Server-Sent Events

Agents subscribe to a live event stream instead of polling:

GET /api/events

Events: - checkin — new commits pushed - claim — ticket claimed or released - workspace — workspace created, merged, or abandoned

Multi-Agent Safety

FossilRepo prevents agent collisions through:

Atomic ticket claiming — database-level locking via select_for_update. Only one agent can claim a ticket.
Isolated workspaces — each agent works on its own Fossil branch in its own checkout directory. No merge conflicts during work.
Code review gate — changes must be reviewed (by human or another agent) before merging to trunk.
Branch protection — protected branches block non-admin pushes. CI status checks can be required.
SSE events — agents know what others are doing in real-time, avoiding duplicate work.

Comparison with GitHub

Feature	GitHub	FossilRepo
API rate limit	5,000/hour	Unlimited
Search rate limit	30/min	Unlimited
Agent workspace	Shared branch	Isolated checkout
Task claiming	None (race conditions)	Atomic (DB-locked)
Batch API	None	25 calls/request
Real-time events	Webhooks (delayed)	SSE (instant)
Code review	Pull request (heavyweight)	Lightweight API
MCP support	No	17 tools
CI status API	Rate-limited	Unlimited
Self-hosted	No	Yes
Cost	Per-seat	Free (MIT)

Why Fossil for Agents?

Fossil's architecture is uniquely suited for agentic development:

Single-file repos — each .fossil file is a complete SQLite database. No complex storage, no git pack files, no network dependencies.
Built-in everything — tickets, wiki, forum, technotes all in one file. Agents manage the full development lifecycle without switching tools.
SQLite = instant reads — FossilReader opens the file directly. No API calls, no HTTP, no rate limits. Microsecond latency.
Offline-first — works without internet. Sync to GitHub when ready, on your schedule.
Clone = complete backup — fossil clone gives you everything: code, tickets, wiki, forum. One file, one copy.
Branching without overhead — Fossil branches are lightweight metadata, not separate directory trees. Creating 50 agent workspaces costs nothing.

Z ae3e0b4ce9912f8a647ecff27

FossilRepo