Agent Frameworks
11 integrations
CrewAI
Python-first multi-agent framework where role-based crews collaborate on tasks, with built-in tools, memory, and process orchestration for production workflows.
Read moreInstructor
Python library that pins LLM outputs to Pydantic models with automatic validation and retries, so you get typed objects instead of raw JSON to parse.
Read moreLangChain
Composable framework for building LLM apps and agents in Python and JavaScript, with retrievers, chains, and tool-calling agents at its core.
Read moreLangChain.js
TypeScript-first framework for building agents and retrieval chains with tools, memory, and a composable retriever interface.
Read moreLangGraph
Graph-based agent runtime for Python and JS with durable state, checkpointing, and human-in-the-loop interrupts for long-running, stateful workflows.
Read moreLiteLLM
A unified API and proxy for calling 100+ LLM providers with one OpenAI-style interface, automatic fallbacks, and per-key budgets.
Read moreLlamaIndex
Python and TypeScript framework for building RAG and agent workflows, with query engines, tool abstractions, and a node-based retrieval pipeline.
Read moreMastra
TypeScript-first agent framework with built-in evals, memory, and workflow primitives for shipping agents that stay fast under real production load.
Read morePydantic-AI
Python-first agent framework that brings Pydantic's type validation to LLM outputs, with structured results, dependency injection, and tool calls you can actually trust.
Read moreVercel AI SDK
A TypeScript-first toolkit for building AI agents with streaming, tool calls, and structured output across every major model provider.
Read moreassistant-ui
React library for building chat and agent UIs with streaming, generative UI, and message primitives, drop a production-grade assistant interface into any React app.
Read moreContext Protocols
1 integration
Memory & State
5 integrations
Editable agent state
Store agent instructions, prompt config, and approved responses as versioned, reviewable content instead of opaque keys in Redis, with edit history and staged rollouts.
Read moreSupabase
Postgres-based backend with auth, realtime, pgvector, and row-level security, the natural home for per-user chat history and ephemeral agent state.
Read moreTurso
Edge-hosted libSQL/SQLite database with embedded replicas and per-tenant branching, fast, cheap storage for chat history and agent session state.
Read moreUpstash Redis
Serverless Redis with a per-request HTTP API for storing chat history, rate limits, and session state in edge and serverless AI agents.
Read moreXata
Serverless Postgres with branching, full-text search, and a typed TypeScript SDK for storing agent state, memory, and structured records.
Read moreVector & Retrieval
5 integrations
Chroma
Open-source embedding database for AI apps, store, embed, and query documents with a few lines of Python or JS, no infra to babysit.
Read morePinecone
Managed vector database for semantic search at scale, with metadata filtering, namespaces, and serverless indexes for production RAG.
Read moreQdrant
High-performance open-source vector database with rich payload filtering, HNSW indexing, and quantization for production-scale similarity search.
Read moreWeaviate
Open-source vector database with built-in hybrid search that fuses BM25 keyword scoring and vector similarity in a single query, fast at scale, schema-light by design.
Read morepgvector
Postgres extension that adds vector columns and similarity search to your existing database, so you can build RAG without a separate vector store.
Read moreLLM Providers
6 integrations
AWS Bedrock
Fully managed AWS service for building and running generative AI agents with foundation models, knowledge bases, and action groups behind a single API.
Read moreAnthropic
Claude's tool-use API lets the model call your functions, but reliability lives in what the tools return: clean schemas, typed results, and content the model can actually trust.
Read moreGoogle Gemini
Google Gemini ships a 1M-2M token context window, but stuffing your whole dataset into the prompt degrades accuracy, inflates cost, and breaks at scale.
Read moreGroq
Groq's LPU inference runs Llama, Mixtral, and other open models at hundreds of tokens per second, so your agent loop replies in well under a second.
Read moreMistral
European open-weight LLM provider with native function calling, structured outputs via JSON schema, and a tool-use loop across Mistral Large, Small, and the open models.
Read moreOpenAI
OpenAI's API gives you function calling, structured outputs, and the Responses API. Here's how to design the tools your agent actually calls so it stops guessing.
Read moreFrameworks & Hosting
5 integrations
Bun
A fast all-in-one JavaScript runtime, bundler, and package manager with a native test runner and TypeScript support, built for low-latency server apps.
Read moreCloudflare Workers
Deploy AI agents to the edge on V8 isolates with sub-millisecond cold starts, no Node runtime, and a global footprint that runs your code close to every user.
Read moreDeno
Secure, TypeScript-first runtime with native permissions, Web-standard APIs, and zero-config deploys to the edge, ideal for small, isolated agent workers.
Read moreNext.js
React framework with App Router, server components, route handlers, and ISR for building and deploying full-stack apps and streaming AI agent endpoints.
Read moreVercel
Cloud platform for deploying Next.js and serverless agent backends, edge functions, streaming responses, and the AI SDK for building LLM apps that ship from a git push.
Read moreObservability & Evaluation
6 integrations
Arize
LLM observability platform with tracing, evals, and span-level analysis for debugging where agents actually fail in production.
Read moreBraintrust
Eval and observability platform for AI apps, with scorers, datasets, and trace logging for measuring agent quality in production.
Read moreHelicone
Proxy-based LLM observability that logs every prompt, response, token cost, and latency across your agent calls with one base-URL change.
Read moreLangSmith
LLM-native observability and eval platform that captures full agent traces, token costs, and tool I/O for debugging production LangChain and LangGraph runs.
Read moreOpenTelemetry
Vendor-neutral observability standard: trace spans, metrics, and logs that show what your agent actually retrieved before it answered.
Read moreWeights & Biases
Track agent runs in W&B Weave, trace every LLM call, tool invocation, and retrieval result so you can find which step breaks before your eval scores tank.
Read more