jmylchreest/aide

Persistent memory, code intelligence, and multi-agent orchestration for AI coding assistants. Works with **Claude Code**, **OpenCode**, and **Codex CLI**.
**Prerequisite:** Bun — the only runtime dependency. The Go binary downloads automatically.
Install
**Claude Code:**
**OpenCode:**
**Codex CLI:**
The Codex plugin provides the MCP server and skills (invoke with $ or /skills); the install step adds lifecycle hooks. On older Codex versions without plugin support, the install step alone configures everything.
The Go binary ships as a per-platform npm package, so it lands on disk with the install (older setups fall back to downloading it on first run). Skills become available immediately.
Install pins the exact version in opencode.json; re-run it with @latest to upgrade. See OpenCode setup for why the pin matters.
What You Get
| Capability | What it does |
| ------------------- | -------------------------------------------------------------------------------- |
| **Memory** | Remembers preferences and context across sessions |
| **Decisions** | Records architectural choices, enforces them in every session |
| **Blueprints** | Seeds best-practice decisions for Go, Rust, GitHub Actions, and more |
| **Code Index** | Fast symbol search, call graphs, and references via tree-sitter |
| **Static Analysis** | Detects complexity, coupling, secrets, and code duplication |
| **Survey** | Maps codebase structure: modules, entry points, tech stack, churn hotspots |
| **Skills** | 23 built-in workflows triggered by natural language |
| **Swarm** | Parallel agents with full SDLC pipelines (design, test, implement, verify, docs) |
| **34 MCP Tools** | Full programmatic access to all capabilities above |
Get Started
Existing project — understand the codebase:
AIDE indexes symbols, discovers modules, tech stack, entry points, and git churn hotspots, then presents the big picture. Full guide
New project — set up guardrails:
Bootstrap with curated best practices, then customise:
Then refine with the decide skill for project-specific choices:
Blueprints cover language idioms, tooling, CI/CD, and common pitfalls. The decide skill handles project-specific architectural choices. Both persist across every future session. Full guide
Share decisions with your team:
Export decisions and memories to git-friendly markdown, then commit them:
Teammates who add AIDE_SHARE_AUTO_IMPORT=1 to .claude/settings.json get them imported automatically at session start. Decisions are append-only per topic (new ones supersede, history kept); memories merge by ULID (newer edits win).
**Teammates without AIDE can still use the folder.** Each .aide/shared/decisions/*.md and .aide/shared/memories/*.md is self-contained markdown with structured frontmatter — any AI assistant can ingest them as context. A short DECISIONS.md / MEMORIES.md explainer is written alongside the exports to document the format. Full guide
Skills
Skills are markdown workflows triggered by keywords. Type naturally — trigger matching is fuzzy.
| Skill | Example Prompt | What Happens |
| -------------------- | --------------------------------------- | ------------------------------------------------------ |
| **swarm** | swarm 3 implement dashboard | Parallel agents with SDLC pipeline per story |
| **plan-swarm** | plan swarm for the dashboard | Decomposes work into stories for swarm execution |
| **decide** | help me decide on auth strategy | Structured decision interview, records choices |