◈
Multi-Agent Support
Run Claude Code, Codex CLI, or Gemini CLI from one interface. Switch agents per session without changing your workflow. Your worktrees, history, and cost tracking work identically across all of them.
CORE
⎇
Git Worktree Isolation
Every session gets its own worktree and branch. Your agent writes to a completely separate directory — zero chance of two sessions stepping on the same file.
CORE
⬡
Parallel Sessions
Run multiple AI agent sessions side by side while each one works from its own isolated branch and directory.
CORE
◎
Session History
Sessions are stored locally as searchable details, timelines, and snapshots, with resume support for interrupted work.
CORE
✓
Integrated Git Workflow
Review diffs, stage files, commit, push, and inspect branch history in the built-in Git window.
CORE
⚙
Remote & VPS Execution
Run sessions on your own VPS — not just your laptop. Connect via device pairing, TLS-secured. Manage local and remote sessions from the same app without SSH juggling.
PRO
📌
Pinned Context Notes
Write sticky project rules once: always use the shared HttpClient, never touch the legacy API. Bonsai keeps them close whenever your agent works.
PRO
📚
Project Knowledge
Bonsai keeps your agent context files — CLAUDE.md, AGENTS.md, GEMINI.md, .cursor/rules and more — in sync across every agent. Edit one, Bonsai detects the drift and offers to apply it everywhere, so the same project rules follow Claude, Codex, and Gemini. Files live on disk, so they work outside Bonsai too.
CORE
$
Usage & Cost Tracking
Fetches /usage after every task. See per-session token cost, daily spend, and efficiency score (tokens per line changed). Set a monthly budget cap.
PRO
✓
Diff Review & Approval
Before your agent commits, see a side-by-side diff and approve or reject. Set auto-commit rules: "auto if only tests changed, ask me otherwise."
PRO
◉
Session Awareness
Live session status, conflict indicators, diff-ready badges, and a unified activity feed across all your agent sessions.
PRO
⚡
Task Queue
Queue tasks across sessions. Bonsai runs them as slots open — in order or in parallel — and keeps the results for review. Dispatch work from the control plane, not from individual terminals.
PRO
◉
Fleet Dashboard
One screen showing every running session — local and remote. Status, cost, branch, task. Open Bonsai and immediately know what is running, what changed, and what needs your attention.
PRO
⊕
Coordinator Sessions
Designate one session as a coordinator that reads fleet state, enqueues tasks, messages other agents, and approves reviews — through normal agent tools, not a hidden control plane. A shared fleet budget caps the whole unit and pauses it cleanly when spent.
PRO
☑
Verifier Sessions
When a session closes, spawn a second agent that reviews the diff against a rubric — tests, lint, task match, secret scan — and posts an advisory pass/fail verdict to the review queue. It never auto-merges; it informs you before you look.
PRO
◆
Agent Recipes
Specialist presets — a provider plus flags, MCP servers, and a system prompt. Pick a test author, doc writer, or design analyst when you start a session. Ships with a starter library you can clone and customise.
PRO
✉
Inter-Session Messaging
Sessions share an inbox and can ask each other questions. Answers flow into the recipient's next prompt automatically — capped and non-blocking, so agents never stall waiting on each other.
CORE