A self-hostable control plane for AI coding agents — and the integrated toolchain it sits on top of.

Warren is the headline project. Point it at a GitHub repo, pick an agent, write a prompt; warren spawns the agent inside a sandbox, streams events back to a live UI, lets you steer mid-run, and pushes the workspace branch when it's done.

• Self-hostable. One container, one volume, one HTTP API, one UI. Run it on a home server, a Fly app, or a cluster.
• Sandboxed per run. Every run gets a fresh isolated workspace via burrow; the host is unreachable.
• Closes the autonomous loop. Polls GitHub for triaged issues, dispatches runs, opens PRs. No external daemon.
• Standalone first. Ships with the claude-code and sapling agents inline — fresh-install only needs a GitHub URL and an Anthropic key.

git clone https://github.com/jayminwest/warren && cd warren
cp .env.example .env
docker compose up -d
open http://localhost:8080

The rest of os-eco is the toolchain warren stands on. Each piece works standalone too, but together they form one coherent agent operating system.

                              Warren  (cloud control plane)
                                 │
                  ┌──────────────┼──────────────┐
                  │              │              │
              Substrate       Context        Runtime
              ─────────       ───────        ───────
              Burrow          Mulch          Sapling
              Plot            Seeds
                              Canopy

Tool	CLI	npm	What it does
Burrow	burrow / bw		OS-isolated sandbox runtime — bwrap on Linux, sandbox-exec on macOS. No Docker, no daemon. Warren embeds it per run; usable standalone for any coding agent.
Plot	plot		Typed, queryable coordination object for multi-agent work — binds together seeds issues, mulch records, prompts, runs, and PRs around a single unit of work. The substrate warren uses to keep humans and agents on the same page.

Tool	CLI	npm	What it does
Mulch	mulch / ml		Structured expertise management — record conventions, patterns, and decisions as you work; retrieve them at the start of every session. In production across multiple engineering teams as the memory layer for their Claude Code / agent workflows.
Seeds	sd		Git-native issue tracking — JSONL storage, dependency graphs, structured plans for LLM decomposition, zero external dependencies. The primary planning and tracking surface for agent work.
Canopy	cn		Prompt management — version-controlled prompt composition with inheritance, pinning, and schema validation.

Tool	CLI	npm	What it does
Sapling	sp		Headless coding agent with proactive context management between every LLM call. An alternative runtime warren can dispatch alongside Claude Code.

Every tool in os-eco shares the same shape:

• Git-native storage — JSONL and YAML files that live in your repo, merge cleanly, and need no external database
• Zero runtime dependencies — each tool is a single Bun binary with no daemon or server
• Multi-agent safe — advisory file locking so concurrent agents don't corrupt shared state
• CLI-first — every operation is a shell command with --json output for scripting and piping
• Consistent UX — shared color palette, status icons, help screen layout, and flag conventions

The headline path is warren. Requires Docker.

git clone https://github.com/jayminwest/warren && cd warren
cp .env.example .env       # add ANTHROPIC_API_KEY + GITHUB_TOKEN
docker compose up -d
open http://localhost:8080

Each tool also stands alone. Install only what you need with Bun >= 1.0:

# Context primitives
bun install -g @os-eco/mulch-cli      # expertise
bun install -g @os-eco/seeds-cli      # issues
bun install -g @os-eco/canopy-cli     # prompts

# Substrate
bun install -g @os-eco/burrow-cli     # sandbox
bun install -g @os-eco/plot-cli       # coordination

# Runtime
bun install -g @os-eco/sapling-cli    # headless agent

Initialize the data layer in a project:

cd your-project
ml init && sd init && cn init
ml --version && sd --version && cn --version

# An agent starts a session
ml prime                              # load project expertise into context
sd ready                              # find unblocked issues to work on
sd update sd-a1b2 --status in_progress

# Agent does work, then wraps up
ml record testing --type convention \
  --description "Always mock fs in unit tests"
sd close sd-a1b2

# Multi-agent coordination via plot
plot init "Add OAuth to billing portal"
plot attach plot-abc1 seeds_issue:sd-a1b2 --role tracks

# Cloud orchestration via warren
# (see warren/README.md for the full UI + API surface)

Overstory (overstory / ov, @os-eco/overstory-cli) is a local-first orchestrator: it spawns agents in tmux + git worktrees, coordinates through a SQLite mail layer, and merges with conflict resolution. Choose it over warren when you want a local-only workflow with no HTTP control plane.

bun install -g @os-eco/overstory-cli
cd your-project && ov init
ov sling sd-c3d4 --strategy worktree     # spawn an agent for an issue

Each tool lives in its own sub-repo with independent git history, CI, and npm versioning. This meta-repo tracks cross-cutting concerns:

os-eco/
  warren/          # sub-repo: @os-eco/warren-cli         — headline control plane
  burrow/          # sub-repo: @os-eco/burrow-cli         — sandbox primitive
  plot/            # sub-repo: @os-eco/plot-cli           — coordination object
  mulch/           # sub-repo: @os-eco/mulch-cli          — expertise
  seeds/           # sub-repo: @os-eco/seeds-cli          — issues
  canopy/          # sub-repo: @os-eco/canopy-cli         — prompts
  sapling/         # sub-repo: @os-eco/sapling-cli        — headless agent
  overstory/       # sub-repo: @os-eco/overstory-cli      — local orchestrator
  branding/        # shared visual spec, CLI standards, checklists
  .mulch/          # ecosystem-level expertise
  .seeds/          # ecosystem-level issues
  .canopy/         # shared prompt templates
  .overstory/      # multi-repo orchestration config

# Run tests, lint, and typecheck for any tool
cd warren     && bun test && bun run lint && bun run typecheck
cd burrow     && bun test && bun run lint && bun run typecheck
cd plot       && bun test && bun run lint && bun run typecheck
cd mulch      && bun test && bun run lint && bun run typecheck
cd seeds      && bun test && bun run lint && bun run typecheck
cd canopy     && bun test && bun run lint && bun run typecheck
cd sapling    && bun test && bun run lint && bun run typecheck
cd overstory  && bun test && bun run lint && bun run typecheck

MIT