AIDE - AI Development Environment

Multi-agent orchestration, persistent memory, and intelligent workflows for AI coding assistants.

Supports Claude Code and OpenCode through a shared core with platform-specific adapters.

Without AIDE	With AIDE
Context lost between sessions	Memories persist and auto-inject
Manual task coordination	Swarm mode with parallel agents
Repeated setup instructions	Skills activate by keyword
No code search	Fast symbol search across codebase
No code quality analysis	Static analysis with 4 analysers
Decisions forgotten	Decisions recorded and enforced

Quick Start

Claude Code

claude plugin marketplace add jmylchreest/aide
claude plugin install aide@aide

OpenCode

bunx @jmylchreest/aide-plugin install

This registers the aide plugin and MCP server. Skills become available as /aide:* slash commands.

Installation

Claude Code - From Marketplace (Recommended)

claude plugin marketplace add jmylchreest/aide
claude plugin install aide@aide

Or register the marketplace in ~/.claude/settings.json (or .claude/settings.json for project-level) so team members are prompted to install:

{
  "extraKnownMarketplaces": {
    "aide": {
      "source": {
        "source": "github",
        "repo": "jmylchreest/aide"
      }
    }
  }
}

OpenCode - From npm

bunx @jmylchreest/aide-plugin install

This modifies your opencode.json to register the aide plugin and MCP server. Check status with bunx @jmylchreest/aide-plugin status, uninstall with bunx @jmylchreest/aide-plugin uninstall.

From Source

git clone https://github.com/jmylchreest/aide && cd aide

# Build (requires Go 1.21+)
cd aide && go build -o ../bin/aide ./cmd/aide && cd ..
npm install && npm run build

# Claude Code
claude --plugin-dir /path/to/aide

# OpenCode
bunx @jmylchreest/aide-plugin install --plugin-path /path/to/aide

The aide Go binary is automatically downloaded when the plugin is installed. No separate binary installation needed.

Permissions (Claude Code)

Add to ~/.claude/settings.json:

{
  "permissions": {
    "allow": [
      "Bash(aide *)",
      "Bash(**/aide *)",
      "Bash(git worktree *)",
      "mcp__plugin_aide_aide__*"
    ]
  }
}

Features

Memory & Decisions

Memories persist across sessions and auto-inject at session start. Decisions record architectural choices that are enforced across all agents.

You: remember that I prefer vitest for testing
# Next session...
You: what testing framework should I use?
AI: Based on your preferences, you prefer vitest for testing.

Use /aide:memorise to store, /aide:recall to search, /aide:forget to clean up, /aide:decide for formal architectural decisions. See Memory System for tagging, scoping, and injection details.

Code Indexing

Fast symbol search using tree-sitter. Supports TypeScript, JavaScript, Go, Python, Rust, and more.

aide code index              # Index codebase
aide code search "getUser"   # Search symbols
aide code symbols src/auth.ts  # List file symbols

Set AIDE_CODE_WATCH=1 for automatic re-indexing on file changes. Use .aideignore (gitignore syntax) to exclude files from indexing and analysis.

Static Analysis

4 built-in analysers detect code quality issues without external tools:

Analyser	Detects
complexity	High cyclomatic complexity functions
coupling	High fan-in/fan-out, import cycles
secrets	Hardcoded API keys, tokens
clones	Duplicated code blocks

aide findings run                         # Run all analysers
aide findings stats                       # Health overview
aide findings list --severity=critical    # Critical findings
aide findings accept --analyzer=clones    # Dismiss noise (hidden from future output)

Findings are searchable via MCP tools during code review and debugging. The file watcher auto-runs analysers on changed files.

Typical workflow: Use /aide:patterns to surface code health issues, then /aide:assess-findings to triage — the AI reads actual code for each finding, accepts noise, and reports what remains actionable.

Status Dashboard

aide status          # Server, watcher, index, findings, stores, env
aide status --json   # Machine-readable output

Modes

swarm 3 implement the dashboard   # Spawns 3 parallel agents with SDLC pipelines
ralph fix all the failing tests   # Won't stop until all tests pass
design the auth system            # Technical spec with interfaces and decisions

See Swarm Mode for SDLC pipeline details, worktree management, and agent coordination.

Skills

Skills are markdown files that inject context when triggered by keywords. Trigger matching uses fuzzy matching for typo tolerance.

Skill	Example Prompt	What Happens
swarm	swarm 3 implement dashboard	Spawns N parallel agents with SDLC pipeline per story
plan-swarm	plan the dashboard work	Decomposes work into validated stories for swarm execution
decide	help me decide on auth strategy	Formal decision-making interview, records architectural choices
design	design the auth system	Technical spec with interfaces, decisions, acceptance criteria
test	write tests for auth	Creates test suite with coverage verification
implement	implement the feature	TDD implementation - make failing tests pass
verify	verify the implementation	Full QA: tests, lint, types, debug artifact check
docs	update the documentation	Updates docs to match implementation
ralph	ralph fix all failing tests	Won't stop until verified complete
build-fix	fix the build errors	Iteratively fixes build/lint/type errors until clean
debug	debug why login fails	Systematic debugging with hypothesis testing
perf	optimize the API	Performance profiling and optimization workflow
review	review this PR	Security-focused code review
patterns	find patterns, code health	Analyze codebase patterns using static analysis findings
assess-findings	assess findings, triage	Triage findings: read code, accept noise, keep genuine issues
code-search	find all auth functions	Search code symbols and find call sites
memorise	remember I prefer vitest	Stores info for future sessions
recall	what testing framework?	Searches memories and decisions
forget	forget the old auth decision	Soft-delete or hard-delete outdated memories
git	help with git rebase	Expert git operations and worktree management
worktree-resolve	merge the swarm branches	Intelligently merges worktrees with conflict resolution

Custom Skills

Create .aide/skills/my-skill.md:

---
name: deploy
triggers:
  - deploy
  - ship it
---

# Deploy Workflow

1. Run tests: `npm test`
2. Build: `npm run build`
3. Deploy: `./scripts/deploy.sh`

Skills are auto-discovered from: .aide/skills/ (project) > skills/ (project) > plugin-bundled > ~/.aide/skills/ (global). Skills are hot-reloaded on changes.

Configuration

Variable	Description
AIDE_DEBUG=1	Enable debug logging (logs to .aide/_logs/)
AIDE_FORCE_INIT=1	Force initialization in non-git directories
AIDE_CODE_WATCH=1	Enable file watching for auto-reindex
AIDE_CODE_WATCH_DELAY=30s	Delay before re-indexing after file changes
AIDE_MEMORY_INJECT=0	Disable memory injection
AIDE_SHARE_AUTO_IMPORT=1	Auto-import shared decisions/memories on start

Reference Documentation

For detailed documentation on all subsystems, see docs/reference.md:

• Architecture - Layered design, hooks, MCP read/write separation
• All 25 MCP Tools - Memory, decisions, state, messaging, code, findings
• CLI Reference - Full command reference
• Memory System - Tagging, scoping, auto-injection, decisions
• Code Indexing - Tree-sitter, file watcher, .aideignore
• Static Analysis - 4 analysers, MCP integration, auto-run
• Swarm Mode - SDLC pipeline, worktrees, coordination
• Storage - File layout, gitignore rules, sharing via git
• Status Dashboard - aide status output details
• Platform Comparison - Claude Code vs OpenCode capabilities
• Quality Guards - Write guard, comment checker, tool enforcement
• Hooks - Lifecycle events and their purposes

Troubleshooting

aide version                              # Check binary
aide status                               # Full system dashboard
AIDE_DEBUG=1 claude                       # Debug logging (or AIDE_DEBUG=1 opencode)

Reinstall:

# Claude Code
claude plugin uninstall aide && claude plugin install aide@aide

# OpenCode
bunx @jmylchreest/aide-plugin uninstall && bunx @jmylchreest/aide-plugin install

Adding Support for New Assistants

See the adapters documentation. AIDE's adapter architecture maps any tool's lifecycle events to shared core functions — skills and MCP tools work out of the box.

License

MIT