aiflow
Turn any repository into a governed, AI-driven software-delivery pipeline with one command — durable task memory, a two-layer code memory (structural graph + semantic RAG), specialist review/audit agents, first-class team collaboration, big token savings, and a real release process.
Version 0.1.1 · MIT License
What it is
aiflow wires Claude Code together with a curated toolchain so an AI agent — or a whole team of humans + agents — can take an issue, plan it, write the code in a consistent style, test it, review it against acceptance criteria, audit it for security and quality, and ship it through a governed branching + release model.
Most people struggle to set up their AI project successfully — especially without deep AI know-how yet. aiflow is built to fix exactly that (start with AI Basics if the terminology is new to you).
The idea: a very good, universal base configuration that works everywhere — because a strong base config beats no config, and “no config” is how most AI-coding projects sadly start. Out of the box it saves roughly 70–80 % of the configuration effort versus using Claude blank. The shipped agents and rules are deliberately generic: customise them to your project (domain language, review focus, test stack) — but even uncustomised they beat a blank start.
The actual goal is production-ready code: reusable, reliable, secure, on current standards; agents that know and respect your architecture, extend it sensibly, push back on requirements that don’t fit it, and propose new layers (caching, search, service seams) where performance demands — plus on-demand reports on accessibility (WCAG), modernisation potential, and security. Token saving remains a goal too, but the quality rules mean it is only partially achieved per task — the real saving is never having to ask twice.
- Token-based & vendor-neutral — your own Anthropic API key or Claude Code OAuth token; git hosts via tokens only, never OAuth.
- Local-first option — run easy work on Ollama models (no key); keep top models for hard reasoning.
- Project-scoped — secrets and settings live in the project (
.env,.aiflow/config.json), never globally. - Cross-platform — Windows, Linux, macOS.
Why teams choose it
| Advantage | How |
|---|---|
| Better memory | Structural graph (graphify) + semantic RAG (cocoindex-code) + durable Beads tasks → the agent looks things up instead of guessing. |
| Fewer tokens | caveman (~75% less output) + rtk (60–90% less CLI noise) + graph/RAG retrieval (~70% vs reading files) + cheap/local routing. |
| Team-ready | Shared Dolt issue DB over your git remote, atomic claiming, pull-before-push. |
| Governed | Conventional Commits, enforced Google style, review gate, security/quality/deps/test/perf/docs audits, branching + releases. |
| Autonomous | The Ralph loop finishes tasks unattended (local, container, or CI). |
Explore the docs
- Getting started — install and build your first project.
- Features — the full capability map.
- Memory: graph + RAG — why aiflow uses both, and how it routes questions.
- Agents — the full roster of delivery, audit, and brownfield agents.
- Models — Claude access, Ollama, context7, adding more models.
- Remote hosts — GitHub, GitLab, Bitbucket, Forgejo, Gitea, custom.
- Team collaboration — many members, one issue graph.
- Configuration — CLAUDE.md, team preferences, custom MCPs.
- Commands — the
aiflowCLI reference. - Workflows & CI/CD — branching models + build/release.
- FAQ · Feedback & contributing
aiflow is glue — huge thanks to the projects it stands on (see Feedback & contributing).
Related topics & tools: Claude Code · Anthropic Claude · AI coding agent · MCP (Model Context Protocol) · Beads issue tracker · Dolt · graphify code knowledge graph · CocoIndex / cocoindex-code semantic code RAG · Context7 · Ollama local LLMs · claude-code-router · rtk · caveman · token optimization · retrieval-augmented generation · agentic software delivery · gitflow · Conventional Commits · GitHub / GitLab / Bitbucket / Forgejo / Gitea.