v0.1 · self-hosted · open source

The modern way
teams ship with AI agents.

Boardroom spawns Claude Code, Codex, and OpenCode in parallel, orchestrates them through visual pipelines, isolates each on its own git worktree, and ships the work back as PRs.

MIT licensed·145 tests passing·AbhiPoluri/boardroom
npm install -g @anthropic-ai/claude-code
fleet · live
5 active
claude · refactor-auth
Extract session token logic into middleware
running
$0.42
codex · spec-tests
Generate vitest specs for billing webhooks
running
$0.18
opencode · ui-pass
Wire empty state into /workspace route
review
$0.61
claude · resolver
Resolve merge conflict in lib/db.ts
running
$0.07
codex · doc-sync
Update API reference for SSE endpoints
merged
$0.23
tokens
847,213
spend
$3.87
PRs open
04
localhost:7391/workspace
live
Good morning, Abhiram.
8 agents running across 4 worktrees. Estimated cost today: $8.42.
Active agents
08+3
PRs open
12+4
Tokens 24h
2.1M+18%
Spend 24h
$8.42+0.6%
fleet activity · 7druns
completedin progress
recentview all
  • claude opened PR #412 — extract session middleware
    2m
  • codex wrote 47 vitest specs for billing
    6m
  • opencode wired empty state for /workspace
    11m
  • claude resolved merge conflict in lib/db.ts
    18m
  • codex synced API reference for SSE endpoints
    24m
built on top of
△Claude Code
◇Codex
○OpenCode
▤MCP
⌥git worktrees
▲Next.js 16
145tests passing
100+commits shipped
20+REST endpoints
3agent runtimes
// the cockpit

Five surfaces, one nervous system.

Each surface in Boardroom solves a specific failure mode of running agents at scale: drift, conflict, cost blowup, and orchestration soup.

Orchestratorlive
orchestrator
›
Routing to claude · codex · opencode...

Type the goal. Watch the fleet form.

Plain-language input. The orchestrator decomposes work into agent tasks and routes between them.

Worktreeslive
worktrees · isolated04
main
head
agent/refactor-auth
+10 / -3 files
agent/spec-tests
+17 / -4 files
agent/ui-pass
+24 / -5 files

Branch isolation by default.

Every agent runs on its own git worktree. No conflicts. No babysitting.

Costslive
cost · last 24h$3.87
claude · refactor-auth
$0.42118k
codex · spec-tests
$0.1847k
opencode · ui-pass
$0.61204k
claude · resolver
$0.0721k
codex · doc-sync
$0.2389k
claude · perf-audit
$0.39133k
claude · refactor-auth
$0.42118k
codex · spec-tests
$0.1847k
opencode · ui-pass
$0.61204k
claude · resolver
$0.0721k
codex · doc-sync
$0.2389k
claude · perf-audit
$0.39133k

Per-agent token & spend tracking.

Live stream of what every agent burned, with breakdowns by run, route, and model.

Auto-PRlive
github · pulls04 open
#412feat: extract session middlewareready
#411test: vitest specs for billingreview
#410fix: empty state for /workspaceready
#409chore: resolve conflict in lib/dbmerged

Commit, push, resolve.

Agents open PRs automatically. When they collide, Boardroom spawns a resolver agent to fix the merge.

▸

Marketplace

100+ skills, MCP server configs, and reusable agent personas. Browse, preview, install.

▸

Cron scheduling

Schedule recurring agent jobs on any cron expression. Daily PR review, hourly site sweep, whatever.

▸

In-browser IDE

File browser, multi-tab editor, syntax highlighting, diff viewer, PR review. Same tab as the fleet.

▸

Cost analytics

Per-agent token tracking and cost breakdown with optimization suggestions baked in.

// pipelines

DAGs you can read at a glance.

Drag, route, evaluate, retry. Hover any node to highlight its dependency path. Click to inspect. Save the workflow and rerun on cron, on push, or whenever.

  • Parallel fan-out across agent runtimes
  • Evaluator nodes with auto-retry on score < threshold
  • Conditional routing on output shape
  • Persisted runs with full I/O replay
workflows·smart-pr-pipeline·5 steps
60%
selected step
1. classifier
Classify the incoming PR. Route to bug-fix or feature-review.
click
new · mcp

Talk to Boardroom from Claude Code.

The Boardroom MCP server exposes the entire fleet API as tools any MCP client can call — Claude Code, Cursor, Windsurf, Claude Desktop. Spawn agents, route work, check costs, all without leaving your editor.

MCP setup guide9 tools · stdio transport · Node 18+
~/myapp · claude code
$claude mcp add boardroom -- node ./boardroom/mcp/dist/index.js
→✓ added MCP server "boardroom"
in your next claude session
›spawn three claude agents to refactor lib/db.ts in parallel
▶ boardroom_spawn_agent × 3
→ claude · split-queries (0.4s)
→ claude · extract-types (0.4s)
→ claude · add-indexes (0.5s)
→spawned 3 agents · isolated worktrees · streaming output
→→ track them at localhost:7391/workspace
boardroom_spawn_agentboardroom_list_agentsboardroom_tell_orchestratorboardroom_list_workflowsboardroom_get_token_usageboardroom_stop_agentboardroom_get_workflow_runboardroom_get_agentboardroom_health
// 60 seconds to fleet

Three steps from prompt to PR.

Boardroom is a self-hosted Next.js app that runs agent processes on your machine. Your Claude login does the auth. No new keys.

terminal
$ npm install -g @anthropic-ai/claude-code
$ git clone github.com/AbhiPoluri/boardroom
$ cd boardroom && npm install && npm run dev
→ ready on http://localhost:7391
  1. 01

    Describe

    Tell the orchestrator what you want shipped, in plain language. No prompt engineering.

  2. 02

    Orchestrate

    Boardroom decomposes the work, spawns agents, and routes output between them.

  3. 03

    Deliver

    Agents commit, open PRs, and report back. You review and merge.

// ship it

Spin up your fleet in under five minutes.

Self-hosted, open source, MIT licensed. Bring your Claude Code login. We bring the orchestration.