Integrations

How Flux talks to other tools — LLM providers, vector stores, frameworks, observability, identity providers, storage, and CI/CD.

Integrations are about the seams: where Flux meets the rest of your stack. Some of these have a built-in Flux module; some are recipes that work because Flux is plain Python and so is the other thing.

Pages here are tiered by how much of the heavy lifting Flux already does.

Sub-groups

LLM providers — Anthropic, OpenAI, Google Gemini, Ollama. Tier 1 across the board. Each provider has a setup page covering API keys, model strings, and the parameters that behave differently from the upstream SDK.

Vector stores — ChromaDB, Pinecone, Weaviate, pgvector. Vector stores plug into long-term memory through flux.tasks.ai.memory. Mostly Tier 2 recipes; pgvector is the closest to Tier 1 because it rides on the Postgres connection Flux already manages.

Frameworks — CrewAI, LangChain, LangGraph, Pydantic AI. They run as ordinary Python objects inside @tasks, so a workflow can drive a framework agent the same way it drives any other library call.

Observability — OpenTelemetry, Prometheus, Grafana. Flux emits traces and metrics when the observability extra is installed and [flux.observability] enabled = true. The page covers exporter setup and the metric/span names Flux produces.

Identity providers (OIDC) — Auth0, Keycloak, Azure AD, Okta, Google. Flux ships a generic OIDC verifier; provider pages cover the discovery URL, audience setup, and role-mapping claims for each.

Storage — S3, GCS, Azure Blob, local filesystem. Output storage backends for large task results. S3 is the primary supported remote; the others are Tier 2/3 recipes.

CI/CD — GitHub Actions, GitLab CI, Docker images. Recipes for running tests, registering workflows, and shipping container builds. All Tier 3.

If you want a recipe that isn’t here, file an issue on the Flux repo — the integrations surface grows by demand.