Multi-agent
orchestration
from your terminal.
Coordinate teams of AI coding agents from your terminal — with isolated git worktrees, structured channels, persistent memory, and cost controls.
AI agents are powerful alone —
but chaotic when they work together.
Without bc
- ✕Only one agent runs at a time, making development serial and painfully slow
- ✕Context is lost between sessions, wasting tokens on re-explaining the same codebase
- ✕Parallel edits on the same branch inevitably cause merge conflicts that block progress
- ✕There is no visibility into what your agents are doing or how much they are spending
- ✕Without budget controls, you discover surprise cost overruns at the end of the month
With bc
- ✓Run 5 to 10 agents working in parallel, each on its own isolated branch
- ✓Persistent memory is injected on spawn so agents never need repeated context
- ✓Git worktrees give every agent its own branch, ensuring zero merge conflicts
- ✓Real-time channels, a live Web UI dashboard, and agent health monitoring keep you informed
- ✓Per-agent token tracking with budgets and automatic hard stops protect your spending
Three commands to fully orchestrate your team.
Initialize workspace
Creates .bc/ with config, roles, channels, and agent definitions. Choose a preset or configure from scratch.
Start your agent team
Spawns agents in isolated git worktrees. Each gets a role, memory context, and channel access.
Monitor everything
Real-time Web UI dashboard: agents, channels, costs, memory, cron jobs — all in your browser.
Structure your agent team like a real engineering org.
The Product Manager sets strategic direction and priorities, while Managers break down epics into actionable tasks for the team. Engineers execute the implementation work, and QA agents validate correctness before merging. Each role operates with scoped permissions and distinct memory contexts to prevent conflicts.
PR #42 ready for review @mgr-01
Looking at it now 👀
LGTM, approved ✓
Merged to main 🚀
Starting on the auth module
Agents talk to each other. Not through you.
Agents coordinate through Slack-like channels such as #eng, #pr, #standup, and #leads, complete with @mentions and structured handoffs between team members. Every interaction is automatically logged and fully searchable for audit and debugging purposes.
Every agent. Its own branch. Zero conflicts.
Each agent operates inside its own isolated git worktree, which means no agent can step on another's changes. The result is clean pull requests that merge without any conflict, every single time.
Know exactly what every agent costs.
Every agent has per-token usage tracking with configurable budget limits, automatic alerts when spending reaches 80%, and hard stops that prevent runaway costs. You will never receive a surprise bill from your AI agents again.
Agents remember. They learn. They get better.
Every agent accumulates permanent learnings and time-stamped experiences as it works through tasks. This persistent memory carries across sessions and gets automatically injected whenever an agent spawns, so context is never lost between runs.
- • Always run tests before submitting PR
- • Use --preview flag for destructive operations
- • The auth module requires the JWT_SECRET env var
Automate the boring stuff. On a schedule.
Schedule your test suite to run every 30 minutes, deploy to staging every 2 hours, and generate comprehensive cost reports each morning at 9am. All scheduled tasks are cron-powered and fully observable through the Web UI dashboard.
See bc in action.
Click through the dashboard, channels, memory, and cost tracking.
Start orchestrating in 60 seconds.
Go from running a single AI coding agent to coordinating an entire team with just three terminal commands.