Orchard Docs

Get started with Orchard

Everything you need to understand Orchard's core concepts and get your team sharing knowledge across AI sessions.

Account & API keys

Create an account — this is the identity your teammates will see in the mesh.

Then add your API keys. Orchard supports Anthropic, OpenAI, and OpenRouter. Keys are stored locally on your machine and never sent to our servers.

Each team member adds their own keys. Orchard does not proxy model calls through a central server — every request goes directly from your machine to the model provider.

Your first project

Point Orchard at an existing repo or start fresh. The engine indexes your codebase and creates a workspace. Then follow these steps:

1
Select seedpacks
Core seedpacks load automatically. Browse the catalog and plant the packs your team needs — each one adds skills, rules, hooks, and tools to every AI session.
2
Choose a theme
Pick from 17+ presets or describe a vibe and let AI generate one. Light, dark, and everything between.
3
Start a session
Open Claude Code, Cursor, Codex, or Windsurf. Orchard is already running in the background — your session loads seedpacks, pulls memory, and inherits your team's knowledge automatically.

Seedpacks

A seedpack is a versioned, SHA256-manifested bundle that gives your AI sessions new capabilities. Think of it as an npm package for your agent's skills, rules, hooks, MCP servers, and CLI tools.

Plant a seedpack and every AI session on your machine gains its capabilities immediately. Share it with your team and propagation completes within 60 seconds.

ComponentWhat it does
SkillsReusable agent capabilities — code review, debugging, test generation, deployment checks
RulesConstraints that shape behavior — coding standards, security policies, naming conventions
HooksShell commands triggered by events — pre-commit validation, auto-formatting
MCP ServersModel Context Protocol extensions — databases, APIs, browser automation
CLI ToolsBundled utilities — scaffolding, migration runners, code generators
TemplatesProject scaffolds and file templates for new code
Shell
$ orchard seedpack list
23 seedpacks installed · 4 core (required) · 19 optional

$ orchard seedpack plant superpowers
Planted superpowers@2.1.0 · SHA256 verified · 12 skills, 4 rules, 2 hooks loaded

The Mesh

The mesh is Orchard's peer-to-peer network. Every laptop running Orchard can join — knowledge, skills, and session memory propagate across devices and teammates within seconds.

Machines discover each other through encrypted tunnels. NAT traversal is built in — no port forwarding, no VPN, no IT tickets. Share a PIN with a teammate and their machine joins your mesh.

The mesh carries metadata only — skills, rules, memory, session summaries, proposals. Never source code. Never API keys. Your repositories stay where they were cloned.

Work offline and the sync layer queues changes locally. Reconnect and queued changes replay automatically. No lockout, no lost work.

TeamSync

TeamSync is the propagation engine inside the mesh. Four GenServers coordinate to push every new learning across the team in under 60 seconds:

GenServerResponsibility
MemorySyncReplicates scoped memory across teammates
PollenClaimerDistributes new skills and rules

Sessions and proposals sync via project-scoped RLS on the existing mirror tables — no dedicated replication workers needed. One engineer commits a new rule — every teammate's agent sees it before the next standup.

Roots discovery

Roots is Orchard's semantic discovery layer. It surfaces what a teammate figured out last month the moment your agent needs it — without anyone typing a question.

A Rust-backed vector index (vecstore) stores embeddings of your team's collective knowledge. When an agent encounters a problem, Roots searches across all team members' memory and session history to find relevant context. The agent already knows where to look.

Roles & permissions

Every team member is assigned one of three roles. Each role inherits the permissions of the roles below it.

RoleCan do
OwnerManage billing, team settings, and everything below
AdminManage seedpacks, proposals, member invites, and everything below
MemberUse all tools, contribute knowledge, submit proposals

Inviting teammates

1
Send an invite
Open the Canopy dashboard and invite a teammate by email. They receive a link to create their account and join your team.
2
They join your workspace
They create their account and add their own API keys.
3
Join the mesh with a PIN
Share the mesh PIN out-of-band (Slack, in person, text). They enter the PIN and their machine joins your team's mesh. Knowledge starts flowing.

Canopy dashboard

Canopy is your team's command center. It runs inside the Orchard Desktop app and shows everything happening across your team's AI sessions.

Active sessions, recent proposals, seedpack changes, team members — all visible in one place. Switch between teams without logging out. Each team has its own projects, seedpacks, proposals, and mesh.

Plugin API

The local engine exposes a versioned, scope-authenticated HTTP API at http://127.0.0.1:19470/api/v1 so a plugin can call Orchard directly -- offline, with no cloud round-trip. It is off by default; enable it in Settings -> Developer (the plugins_api.enabled flag).

Authentication

Mint a scoped key in Canopy (Settings -> Developer -> API keys) or with orchard keys create. Send it on every request as a Bearer token (or the X-Orchard-Api-Key header). The plaintext key is shown once at creation -- store it securely.

Shell
curl -H "Authorization: Bearer orch_..." \
  http://127.0.0.1:19470/api/v1

Scopes

Keys carry a closed set of scopes and each endpoint enforces one -- request only what you need. The live vocabulary and preset tiers are served at GET /api/v1/scopes.

ScopeGrants
projects:read / projects:writeRead or manage projects
terminal:read / terminal:spawn / terminal:writeRead, spawn, and drive sessions
proposals:read / proposals:write / proposals:executeRead, author, and build proposals
files:read / files:writeRead and write within a bound project

Endpoints

The curated /api/v1 surface covers discovery, projects, sessions and terminals, proposals, files, and events. The full machine-readable contract is the OpenAPI 3.1 document at GET /api/v1/openapi.json.

Using the client

The @orchard-ai/client TypeScript SDK resolves the engine URL and attaches your key automatically.

TypeScript
import { OrchardClient } from "@orchard-ai/client";

const client = new OrchardClient({ apiKey: "orch_..." });
const info = await client.discover();              // GET /api/v1
const projects = await client.request("GET", "/api/v1/projects");

Common questions

Do I have to change how I work?
No. Keep using Claude Code, Codex CLI, Grok Build CLI, OpenCode, or Hermes Agent exactly as you do now. Orchard watches your sessions, syncs knowledge, and keeps the right context available without per-tool setup.
Does my code leave my machine?
No. The engine runs locally at localhost:19470. TeamSync carries metadata only — skills, rules, memory. Never source files. Your repositories stay where they were cloned.
Does Orchard compete with Claude, OpenAI, or the other model providers?
No — Orchard sits on top of them. Think of it as the team layer for the tools you already use. When one teammate gets a great answer from a model, that knowledge lives in your seedpacks and helps every teammate the next time they start a session. Behind the scenes, Orchard watches each session, distills the durable lessons into shared rules, skills, and memory, and syncs them to every teammate machine in under a minute. You spend fewer tokens on questions your team has already answered, and every session starts with more context. The model providers ship the intelligence; Orchard makes sure your team only has to learn each lesson once. Your data stays yours — it lives on your machines, in your team, under your control.
Which AI tools does it work with?
Claude Code, Codex CLI, Grok Build CLI, OpenCode, and Hermes Agent — simultaneously. Orchard keeps shared memory, skills, and rules available so each tool starts with better context.
Does it work offline?
Yes. The engine and all tool integrations run locally. TeamSync queues changes and catches up on reconnection.
How does knowledge reach my teammates?
MemorySync and PollenClaimer push new knowledge across the mesh in under 60 seconds. Sessions and proposals sync via project-scoped RLS on the existing mirror tables. Atomic. Conflict-free.
What happens if we cancel?
Everything stays on your machines. TeamSync stops replicating. Seedpacks, memory, and rules remain local and readable. No data held hostage.

For the terminal

The orchard CLI drives the same workspace from your shell.

The CLI talks to the same local engine as the desktop app, so people, scripts and the agents themselves can use it. Install it in one line and manage sessions, proposals and context from the terminal.

orchard
  1. npm install -g @orchard-ai/cli# one self-contained binary
  2. orchard sessions list# list local agent sessions
  3. orchard proposals list# tracked plans and builds
  4. orchard dispatch run "<prompt>"# launch a task in a sibling session
  5. orchard api /api/help# discover the local engine API

Install in one line

npm install -g @orchard-ai/cli
  • One self-contained binary, with no Python or pip to install.
  • The npm installer downloads the binary for your platform from the latest cli/v* release and verifies it against SHA256SUMS.
  • Builds for macOS (Apple Silicon and Intel), Windows x64 and Linux x64 ship with every tagged release.
@orchard-ai/cli on npm