ReqLLM

Join the community! Come chat about building AI tools with Elixir and coding Elixir with LLMs in The Swarm: Elixir AI Collective Discord server.

A Req- and Finch-backed package to call LLM APIs that standardizes requests and responses across providers.

Why Req LLM?

LLM APIs are inconsistent. ReqLLM provides a unified, idiomatic Elixir interface with standardized requests and responses across providers.

Unified architecture:

• High-level API – Vercel AI SDK-inspired functions (generate_text/3, stream_text/3, generate_object/4 and more) that normalize requests and responses across providers.
• Provider transports – Req powers request/response calls; Finch powers streaming. Dedicated provider modules or the shared catalog gateway translate model metadata, options, bodies, and responses behind the same public API.

Provider Integrations and Model Support

ReqLLM gets model metadata from LLMDB. It sends a request through one of two integration paths:

• Dedicated provider module: A registered ReqLLM module owns the provider's authentication, routes, wire format, and special behavior.
• Shared catalog gateway: The provider has no dedicated ReqLLM module. LLMDB supplies an execution contract, and ReqLLM sends the request through its shared OpenAI Chat Completions, Cohere-compatible rerank, or decision evaluation adapter.

These paths describe request routing. They do not describe support quality. A model can be in LLMDB before ReqLLM has verified its API behavior. Support tiers such as First-class describe fixture evidence for an exact model and operation. See Model Support Evidence for that detail.

Community contributions can add either integration path. After a provider-specific module is merged, it is a dedicated provider module. The origin of the contribution does not change its integration type.

Hosted APIs With Dedicated Modules

Each provider in this table has a registered module in ReqLLM. Main operations show the module's broad scope. The selected model can support fewer operations.

This is the beginning of the README. Read the rest at the original. Read the full README ↗