Bodhiorchard is the open-source reference implementation of Agent-Driven Development (ADD) — a software development methodology where twelve specialised AI agents handle the busywork (specs, estimates, test plans, triage, retrospectives) and humans keep the decisions that matter. It works as a self-hosted Jira alternative for the full lifecycle: intake → spec → design → development → testing → deploy → retrospective. The data plane stays on your hardware; inference runs on Claude Code today, with Ollama and OpenAI next.

Less process. More shipped. The full methodology lives at bodhiorchard.ai.

Quick Start

The fastest path — no clone, no build. Needs only Docker (running) and Node.js 18+:

npx bodhiorchard init     # detect ports, generate secrets, pull prebuilt images
cd bodhiorchard
npx bodhiorchard start    # start the stack and open the setup wizard

The wizard opens at http://localhost:3000/setup — finish there (organization, admin account, AI provider, repositories). bodhiorchard stop brings it down; your data volumes persist.

Prefer to build from source?

git clone https://github.com/mickyarun/bodhiorchard.git
cd bodhiorchard
docker compose up

Already running claude on your laptop? Hybrid mode reuses your Claude Pro/Max login instead of an API key:

npm install && npm run setup && npm run dev

→ All three paths, the full bodhiorchard CLI command list, prerequisites, mode-switching, and a ready-made demo dataset: Deployment Guide

Why Bodhiorchard

Old way vs Agent-Driven Development

Screenshots

The gamification layer rebuilds developer skill profiles nightly — skills compound, badges unlock, and the leaderboard reflects shipped value, not ticket count:

More walkthroughs: Setup · Slack triage & MCP tools · Requirements & estimation · Design phase · Development & retrospective

How it works

Every feature lives in one BUD (Business Understanding Document) that moves through the lifecycle, with an agent driving each phase and a human gate between phases:

bud → design → tech_arch → development → testing → uat → prod → closed
                                                              ↳ discarded (any time)

• Slack-native intake — post in #feature-requests, the Triage Agent interviews you in-thread, checks duplicates via vector search, and queues for PM approval.
• Multi-repo code intelligence — a tree-sitter code-graph indexer scans your repos, detects feature clusters across them, and answers blast-radius queries ("what breaks if I change this?").
• Smart quality loops — production bugs auto-link to the originating BUD; threshold breaches trigger reassignment; every fix feeds the knowledge base.
• Developer skill profiling — per-module expertise scores rebuilt nightly from git history, with bus-factor alerts and best-fit assignment.
• Living knowledge — git repos, agent skills, the central DB, and vector search stay auto-synced and feed every agent prompt.
• Enterprise-grade security — multi-tenant isolation, RBAC with 9 roles, AES encryption at rest, JWT auth, full audit trail.

The Twelve Agents

Every agent action is audited and reviewable in the UI. Triggers, hand-offs, and capabilities: AGENTS.md.

The Manifesto

Eight principles, deliberately echoing the Agile Manifesto's "X over Y" structure:

Architecture

Stack: FastAPI + SQLAlchemy 2.0 (async) on Python 3.12 · Vue 3 + Vuetify + PlayCanvas (3D) · Colyseus multiplayer · PostgreSQL 16 + pgvector · Redis · local ONNX embeddings (fastembed).

backend/     FastAPI — REST API, 12 agents, MCP server, code-graph indexer
frontend/    Vue 3 SPA — BUD board, Living Tree 3D dashboard, settings
multiplayer/ Colyseus server — shared 3D world state
shared/      World layout consumed by frontend + multiplayer
examples/    TaskFlow demo repos (cross-repo feature detection)

The agent layer is engine-independent: Claude Code runs the codebase-aware agents, the Anthropic direct API handles lightweight ones, and Ollama / OpenAI / Codex are next — adding an engine is API rewiring only. Bodhiorchard also runs its own MCP server so Claude Code (or any MCP client) can drive the BUD lifecycle, query the code graph, and read team context.

Documentation

FAQ

What is Agent-Driven Development?

A methodology where specialised AI agents — not humans running status meetings — drive every phase of the software lifecycle, while humans review, decide, and steer. It sits in the slot Agile occupied in 2001: a reorganisation of work around new capabilities. The methodology is free to adopt with or without Bodhiorchard — see bodhiorchard.ai.

Is Bodhiorchard a self-hosted Jira alternative?

Yes — for the full workflow layer (intake through retrospective). It's especially relevant for teams affected by Atlassian sunsetting self-hosted Jira licences (new licences end March 2026, full shutdown 2029).

Does my code leave my machine?

The data plane — Postgres, embeddings, BUDs, scanned repos, audit log — is always local. Only LLM prompts leave, and only when you choose cloud inference. Fully air-gapped operation arrives with the Ollama preset.

Do I need a Claude subscription?

No. Either paste an Anthropic API key (pay-per-token) or, if you do have Claude Pro/Max, run Hybrid mode and the agents inherit your flat-rate claude login session.

How is this different from Cursor, Tabby, or Continue?

Those are IDE-side coding assistants for an individual developer. Bodhiorchard is the orchestration and project-management layer above them — intake, specs, estimation, assignment, test plans, bug linking, retrospectives. Pair them freely.

How does it compare to GitHub Copilot Workspace?

Copilot Workspace is cloud-only and GitHub-bound. Bodhiorchard is self-hosted, repo-agnostic (GitHub, GitLab, self-hosted Git), and AI-engine-agnostic.

What does "BUD" mean?

Business Understanding Document — one markdown document per feature holding spec, tech spec, test plan, acceptance criteria, and full history.

Can I use it commercially?

Yes — Apache 2.0, including embedding in proprietary products. Contributions require DCO sign-off (git commit -s). Commercial licences with support are available separately from the maintainer.

Does it run on Windows?

Yes, via WSL2 in either deployment mode.

Roadmap

• Shipped — core platform (auth, multi-tenant, RBAC), BUD lifecycle, feature registry with vector search, repo scanning + code intelligence, skill profiling, 3D Living Tree, Slack-native triage and real-time Slack bot conversations, GitHub webhook pipeline (PR-merge → BUD status + repo re-scan), automated test-plan generation from BUDs (unit / integration / e2e / security / UAT), CI/CD integration (QA-automation repo wired via MCP; test commits flow back to the BUD), MCP server, 12 agent definitions, npx bodhiorchard installer
• Next — autonomous agent execution engine: small, low-risk BUDs run tech spec → code → review → test end to end, with a human approval gate before release (deliberately not lights-out)
• Future — multi-org marketplace, custom agent builder, analytics dashboards, mobile companion, plugin ecosystem

Contributing

Contributions welcome — fork, branch, lint (ruff / vue-tsc), and open a PR. All commits need DCO sign-off (git commit -s); see CONTRIBUTING.md.

Why "Bodhiorchard"?

Bodhi (Sanskrit: "awakening") is what the Buddha attained under the Bodhi tree. An orchard is a grove tended with intention so it can bear fruit. The software industry builds tools to make life better, yet the process of building them consumes our lives — ceremonies, status updates, and busywork that were meant to help but became the work itself.

Bodhiorchard exists because AI should give humans their time back. The agents are the orchardists: they water, prune, and tend the soil so the trees can grow and the humans who planted them can step back and enjoy the harvest.

Build well. Then go outside.

License

Apache License 2.0 — see LICENSE and NOTICE.