Documentation
Everything you need to turn requirements into reliable software — install in one command, then run the reliability loop: specify, ground, build, remember, verify.
Quick Start
Install the platform, or pick the piece that fits the job. Works on a Claude subscription or an API key. Everything is open source under Apache 2.0.
attune-ai
The whole platform: spec engine, AI workflows, project memory, retrieval grounding, and verification. 23 workflows, 27 skills, 60 MCP tools.
attune-rag
Lightweight, LLM-agnostic RAG pipeline with pluggable corpora. Grounds generated content in your actual source. Also ships inside attune-ai.
attune-verify
Generation fact-checker for the attune family. Verifies the named entities in LLM output — imports, CLI flags, links — so hallucinations fail before they ship.
Plugin
Install from the marketplace. Type /coach for progressive help right in your terminal.
Companion packages
attune-help is the lightweight reader for .help/ directories — 1 dependency, no AI key required. Its Claude Code plugin lives in the same Smart-AI-Memory/attune-ai marketplace as attune-ai itself. attune-author's AI authoring tools have been consolidated into attune-ai; the standalone package is archived, and all released versions stay installable.
Attune AI Platform
Beyond AI workflows and verification, the platform keeps your docs grounded too: scan your codebase, generate templates, detect when code drifts, and regenerate stale content.
1. Bootstrap
Scan your project to discover features. The scanner reads modules, classes, and functions, then proposes a features.yaml manifest.
2. Generate
For each feature, three templates are created: concept (what is it?), task (how to use it), and reference (full API detail). Content comes from actual source code.
3. Staleness Detection
Source file SHA-256 hashes are stored in template frontmatter. When code changes, stale templates are flagged automatically.
4. Maintenance
Regenerate only the templates whose source changed. Hand-written templates are preserved. Run on-demand or via CI.
Attune Help (Standalone Reader)
A lightweight Python package that reads .help/ templates. 1 dependency, 6 files, embeddable anywhere.
Features
- 11 dependency — python-frontmatter. No bloat.
- 2Progressive depth — concept on first ask, task on repeat, reference on third.
- 3Session storage — file-based or custom backend. Tracks depth per topic.
- 4Multiple renderers — plain text, CLI, Claude Code, marketplace.
Usage
engine = HelpEngine(
template_dir=".help/templates"
)
result = engine.lookup("security-audit")
Claude Code Plugin
Install from the marketplace. Progressive help, project bootstrapping, and 27 skills right in your terminal.
Install
/coach Commands
/coach <topic>Progressive lookup. First call returns concept, repeat returns task, third returns reference.
/coach initBootstrap a .help/ directory for your project. Scans source code and generates features.yaml.
/coach statusCheck template freshness. Shows which features have drifted from their source files.
/coach maintainRegenerate stale templates. Only updates files whose source hashes changed.
Workflows & Skills
The build half of the loop: 23 workflows, 27 auto-triggering Claude Code skills, and an MCP server with 60 registered tools — review, tests, bug prediction, refactor, and release prep.
Key Workflows
25 Claude Code Skills
Skills auto-invoke from natural language. Type the topic and the right skill fires.
Run from CLI
Frequently Asked Questions
What is attune-ai?
A development platform built around project memory that turns requirements into reliable software. Memory leads; AI workflows, retrieval grounding, and verification support it in one reliability loop: specify with /spec, ground every change in your real code, build with a team of workflows, remember what worked across sessions, and verify the output before it ships.
How does it keep generated content from drifting?
Retrieval grounding (powered by attune-rag, a built-in dependency) keeps generated content anchored to your actual source — mean faithfulness is at least 0.97 and CI-gated, so drift fails the build. For docs specifically, templates carry source file hashes; when code changes, staleness detection flags the affected templates so they can be regenerated automatically.
Do I need attune-ai to read templates?
No. The standalone attune-help package (1 dependency) can read any .help/ directory without the full framework. Generate templates with attune-ai, then ship them alongside attune-help for a minimal runtime with no Anthropic API key required.
Can I write templates by hand?
Yes. Hand-written templates are preserved during regeneration. Remove the auto-discovered tag from frontmatter and the maintenance system will skip the file.
How does staleness detection work?
Each template stores SHA-256 hashes of the source files it was generated from. The maintenance workflow re-hashes those files and compares. If the hash changed, the template is flagged stale and queued for regeneration.
Is it free? What do I need to run it?
Released under the Apache License 2.0 — free for personal, commercial, and enterprise use, with full source access.
What Claude Code skills are included?
26 auto-invoking skills: security audit, smart test, code quality, bug prediction, doc generation, refactor planning, release prep, planning, spec-driven development, fix-test, workflow orchestration, RAG-grounded code generation, content verification, cross-session recall, memory and context, personal memory, image analysis, batch processing, capability catalog, discovery sweep, form-driven elicitation, the coach help system, the attune hub router, single-source feature-page authoring, cross-model diff review, and the multi-LLM round table.
Where do I install attune-help from?
PyPI: `pip install attune-help` — the standalone reader, no API key required. Its Claude Code plugin lives in the same Smart-AI-Memory/attune-ai marketplace as attune-ai itself: `claude plugin marketplace add Smart-AI-Memory/attune-ai`, then `claude plugin install attune-help@attune-ai`. (The old Smart-AI-Memory/attune-docs marketplace is retired.)
What happened to attune-author?
Its AI authoring capabilities were consolidated into attune-ai in July 2026, and the standalone package is archived. Already-released versions stay installable from PyPI, but no further releases are planned — for template authoring, install attune-ai and use the author-feature skill or /coach maintain.
Ready to Get Started?
Install in one command, then run /spec on your next feature and turn requirements into reliable software.