Cognis umbrella crate: agent builder, multi-agent orchestration, memory, middleware (rate limit, retry, PII, prompt caching), built-in tools, and re-exports of cognis-core, cognis-graph, cognis-llm, and cognis-rag.
LLM client and provider abstractions for Cognis: Client, LLMProvider trait, chat options, tool definitions, and streaming. Provider implementations (OpenAI, Anthropic, Google, Ollama, Azure) are feature-gated.
Unix native interface for LLMs
Foundation traits and types for Cognis: typed Runnable<I, O>, Message, errors, streaming, prompts, output parsers, callbacks. Zero external integrations.
Stateful graph engine for Cognis: typed Graph<S>, Pregel-style superstep executor, per-field reducers, interrupts, time-travel via checkpointers (in-memory, SQLite, Postgres), and streaming.
RAG primitives for Cognis: embeddings, vector stores (in-memory, FAISS, Chroma, Qdrant, Pinecone, Weaviate), retrievers, text splitters, document loaders, and incremental indexing pipelines.
Procedural macros for Cognis: #[tool] attribute for tool definitions and #[derive(GraphState)] for graph state with per-field reducers.
Pluggable observability for Cognis: bridges CallbackHandler events to Langfuse and (on the roadmap) OpenTelemetry.
Batteries-included agent framework built on cognis and cognisgraph
State graph orchestration framework for multi-actor agent workflows
Typed state machine framework for LLM agents - Rust implementation of LangGraph
Checkpoint persistence for Juncture state machine executions
No description provided.
No description provided.
No description provided.
No description provided.