v8.5.0Spec-driven development platform

Build software from specifications, not prompts.

Attune-AI gives your AI coding agent a spine: write a spec, ground every change in your real code, carry what worked into the next session, and verify the output before it ships. Workflows, memory, retrieval grounding, and verification — one platform for Claude Code.

attune-ai on PyPI Star on GitHub See How It Works

21,386 tests | 85% coverageApache 2.0

One requirement, end to end

Specify

Socratic spec engine: requirements, design, and tasks with an approval gate.

Ground

RAG retrieval cites your code so the agent doesn't invent APIs.

Build

17 multi-stage workflows: review, tests, bug prediction, refactor.

Remember

Cross-session memory and a lessons corpus surface what worked before.

Verify

Fact-check generated content: imports, flags, links, counts — all real.

The problem

AI generates code.
Projects require knowledge.
Specifications drift.
Documentation gets stale.
Decisions disappear.

The solution

Attune keeps software aligned with its requirements by combining five capabilities — each one a job, not a buzzword:

Spec-driven development keeps requirements from drifting.
Project memory preserves decisions across sessions.
Grounded retrieval anchors outputs to your real code.
Synchronized docs stay current as the code changes.
Verification reduces hallucinations before they ship.

The Platform

Four pillars, one outcome

Each pillar is a real, shipped capability — not a roadmap promise. Together they keep your AI agent's work grounded, remembered, and verified.

AI workflows

Specialist teams, not one prompt

17 multi-stage workflows run teams of 2–6 Claude subagents to review code, surface vulnerabilities, generate tests, and plan refactors — with cost-tiered model routing.

Security audit, code review, bug prediction, release prep
Cheap / capable / premium model routing
Structured, readable reports

Project memory

Your agent stops starting from zero

Findings from each session are stashed and recalled in the next. A retrievable lessons corpus surfaces the right engineering lesson at the moment a prompt needs it.

Local-first by default — no cloud required
Optional Redis semantic tier (local Ollama embeddings)
Automatic recall, or on demand with /recall

Retrieval grounding

Answers anchored to your code

Keyword + semantic retrieval keeps generated content grounded in your actual source. Mean faithfulness ≥ 0.97, CI-gated — drift fails the build.

Powered by attune-rag — built in, no extra install
Citations back to source
Faithfulness measured, not assumed

Verification

Catch hallucinations before they ship

Fact-check LLM output against source-of-truth: confirm imports import, CLI flags are real, links resolve, and counts match — before the change reaches main.

Verifies docs, code, and generated content
Closes the loop the spec opened
Built from the discipline that runs this project

multi-stage workflows

MCP tools

auto-triggering skills

≥0.97

RAG faithfulness (CI-gated)

100%

open source · Apache 2.0

Flagship package · standalone

attune-rag — grounding you can use anywhere

The retrieval engine inside attune-ai is its own package. Lightweight, LLM-agnostic RAG with pluggable corpora — works with Claude, Gemini, or any LLM. Built into attune-ai for grounding, and equally happy standing on its own in any Python project.

pip install attune-rag

Visit attune-rag.dev attune-rag on PyPI

LLM-agnostic

Claude, Gemini, or any model — not locked in

Faithfulness ≥ 0.97

CI-gated — drift fails the build

Pluggable corpora

Point it at docs, code, or any source

Standalone or built-in

Drop into any project, or get it free with attune-ai

Semantic memory

Powered by Redis when you want it

Memory is local-first by default — nothing leaves your machine. Plug in Redis Agent Memory Server for semantic search over your project memory: local Ollama embeddings, int8 vector quantization, no cloud required.

pip install 'attune-ai[redis]'

Local-first

Default backend is on-disk — no service required

Redis AMS tier

Opt in for semantic recall across sessions

Ollama embeddings

Runs locally — no API key, no cloud

int8 quantization

Compact vectors, shipped in attune-ai 7.4

Sister Family

The Documentation Toolchain

attune-ai began as a tool to help people work with AI, then grew into an engineering toolkit focused on Claude — and later Redis. The documentation toolchain came after, built with the same development discipline, and split off as four standalone packages forming an end-to-end author → reader loop. Use the full stack, or drop in just the piece you need.

1
attune-author
Generates 15 kinds of source-grounded templates with per-type LLM polish. Runs at dev time or in CI.
2
attune-rag
Keyword + semantic retrieval, mean faithfulness ≥ 0.97 (CI-gated) — answers stay grounded in your code.
3
attune-help
Reads the templates at runtime. 1 dependency, no API key required. Embed in any Python tool.
4
attune-gui
Local dashboard. Browse templates, edit specs, run commands — one pane for the whole stack.

All four are open source — Apache 2.0.

author_reader_loop.py

# 1. attune-author — generate polished, source-grounded
#    templates from your codebase (CI or dev time)
from attune_author.generator import generate_feature_templates
generate_feature_templates(feature, help_dir=".help", project_root=".")

# 2. attune-rag — keyword + semantic retrieval keeps
#    answers grounded. Mean faithfulness ≥ 0.97, CI-gated.

# 3. attune-help — read them at runtime. 1 dependency,
#    no API key required. Embed anywhere.
from attune_help import HelpEngine
engine = HelpEngine(template_dir=".help/templates")
print(engine.lookup("security-audit"))

# 4. attune-gui — local dashboard tying it all together.
#    pip install attune-gui && attune-gui --open

Core Capabilities

Six Ways to Use It

The attune-ai framework, its Claude Code plugin, and the four-package documentation toolchain we built with it. Pick the piece that fits the job.

attune-ai

Full framework. Workflows, staleness detection, MCP server, and 17 auto-triggering skills for Claude Code.

attune-help

Lightweight reader. 1 dependency, 6 files. Embed progressive help in any CLI tool, notebook, or internal app.

attune-author

AI authoring companion. Generates 15 kinds of source-grounded templates with per-type polish prompts.

attune-rag

Keyword + semantic retrieval over your Markdown corpus. Mean faithfulness ≥ 0.97, CI-gated — answers stay grounded.

attune-gui

Local dashboard. Browse templates, edit specs, run commands, and watch jobs — one pane for the whole stack.

Claude Code Plugin

Type /coach in Claude Code. Progressive help in your terminal — no setup required.

Ship software you can trust your agent built

Install from PyPI, run /spec on your next feature, and let the platform keep the work grounded, remembered, and verified.

Read the Docs Star on GitHub

🔄Auto-Freshness

Source hashes detect code drift. Stale templates regenerate automatically — your docs stay in sync with your codebase.

✏️Human-Enhanceable

Edit generated templates freely, or write from scratch. The engine respects hand-written content and never overwrites your work.