Skip to content
Factflow

Factflow

Declarative, queue-driven AI data pipelines. YAML in, vector-indexed knowledge out.
Get started CLI reference GitHub

Where to go

Section titled “Where to go”

For operators

Install the CLI, submit a pipeline, watch an execution run, replay failed stages.

Install the CLI →

For integrators

91 HTTP endpoints across 14 tags. Live /openapi.json, plus a per-endpoint reference.

API reference →

For code readers

Every backend package, its public API surface, and what it’s for.

Package reference →

Core ideas

Section titled “Core ideas”
Pipelines are YAML Declarative routes, adapters, conditions. Operators iterate without a deploy.
Queue-scoped per execution Multiple executions share one broker without cross-talk. ExecutionScopedQueue handles the prefixing.
Storage is the replay source of truth Every artefact is persisted with a metadata sidecar. Replay reads from storage — not a cache.
Commit-independent lineage One row per message per adapter. A pipeline failure still records lineage; a lineage failure never blocks the pipeline.

The stack

Section titled “The stack”

Backend. Python 3.13+, FastAPI, uv workspace. 16 packages: one core protocol package, seven shared services (config, lineage, LLM, infra, engine, execution, server), eight workflow packages (webscraper, markdown, embeddings, boost, translator, knowledge, sharepoint, replay).

CLI. Go 1.26+, cobra. Covers every API endpoint. Installed via the radicalAI Homebrew tap.

Frontend. React 19 + TypeScript + Vite. Consumes the same API the CLI does.

Infrastructure. PostgreSQL 18 with pgvector for embeddings + full-text. Artemis (default) / RabbitMQ / Pulsar for messaging. Filesystem / MinIO / S3 for object storage.

Section titled “Links”