Agents solved coding. Planning became the critical path. SpecFlow turns your scattered specs into one shared knowledge graph. People and coding agents build from the same, complete context.
With AI coding agents, implementation takes a fraction of the time it used to. Planning did not get faster. Even with AI, writing a solid spec still takes weeks, and agents build exactly what an unclear spec says.
A real spec is dozens of documents, decks and Drive folders, written by different people at different times. Requirements repeat and contradict each other, and key decisions are missing. Teams find out while they build, which is the most expensive moment.
The graph stays current: when a document changes, the next build updates only what that change affects. Its vocabulary comes from an ontology file, so you can fit it to your domain without code changes.
make env-local # writes local secrets and asks for your LLM gateway settings
make up WITH=temporal # starts the whole stack in DockerOpen http://127.0.0.1:3001. Prerequisites, the architecture and the development commands are in QUICKSTART.md.
- QUICKSTART.md: run it, how it fits together, develop
- docs/configuration.md: every setting
- docs/graph/: how the knowledge graph is built; docs/graph/plan/: implementation plans on the graph
- docs/file-parsers/: how uploads and Drive files become Markdown
- Rosetta: the harness that turns plan steps into reviewed code
Contributions are welcome. See CONTRIBUTING.md and our Code of Conduct. To report a vulnerability, follow SECURITY.md.