# Alternatives to AutoGen > Source: https://theagentsindex.com/microsoft-autogen/alternatives (current datasheets, evidence-ranked) Candidates share the baseline category or are previously researched peers. Requirements, preserved strengths, explicit trade-offs and research coverage decide the order; paid tier does not. ## Shortlist ### 1. Agno Python SDK, self-hosted AgentOS runtime and hosted control plane for building and operating agent platforms. **Research coverage:** 35/35 criteria **Where it improves on AutoGen:** - [Debugs and fixes: yes — Bug fixing is a documented pattern: reproduce, patch, rerun the failing tests.](https://docs.agno.com/use-cases/coding-agents) - [Reviews code: yes — Code review over a diff is one of the documented coding-agent patterns.](https://docs.agno.com/use-cases/coding-agents) - [Done-for-you service: partial — You run the software yourself, but Enterprise adds vendor engineers on the build: dedicated help with your product agent, monthly office hours and ChatGPT/Claude app help.](https://www.agno.com/pricing) **Trade-offs against AutoGen:** - [Open source: partial — The SDK and AgentOS runtime are Apache-2.0 on GitHub, while the control plane is a hosted product whose self-hosted variant is quoted per deal on Enterprise.](https://github.com/agno-agi/agno) [View Agno](https://theagentsindex.com/agno) ### 2. AgentsKit Composable TypeScript packages for agent runtime, tools, memory, RAG and headless chat UI. **Research coverage:** 34/35 criteria **Where it improves on AutoGen:** - [Reviews code: yes — Two review personas ship with the skills package, one reviewing diffs for bugs and security and one checking a diff against project conventions, and the registry offers an installable code-reviewer agent.](https://www.agentskit.io/llms.txt) - [Claude: yes — A native Anthropic adapter covers the Claude family with streaming, tool calls and vision, and Claude also reaches the runtime through the Bedrock adapter.](https://www.agentskit.io/llms.txt) - [Open models: yes — Local and open-weight inference is a documented path, with adapters for Ollama, LM Studio, vLLM and llama.cpp alongside hosted open-model services such as Together and Fireworks.](https://www.agentskit.io/llms.txt) **Trade-offs against AutoGen:** - [In a browser: partial — Browser chat is available as headless bindings for seven frameworks that you host in your own app; the vendor runs no hosted web product for the library itself.](https://www.agentskit.io/docs/compare) [View AgentsKit](https://theagentsindex.com/agentskit) ### 3. Google Agent Development Kit (ADK) Open-source framework for building, evaluating and deploying AI agents in Python, TypeScript, Go, Java or Kotlin. **Research coverage:** 34/35 criteria **Where it improves on AutoGen:** - [Ships it: yes — Deployment is part of the framework: one command pushes an agent to Google Cloud, and the same agent can be containerised and run on infrastructure you control instead.](https://adk.dev/) - [Agent permissions: yes — Autonomy is set per tool by the developer: a tool can be configured to pause and ask a human or a supervising system before it runs, though the vendor labels this confirmation feature experimental.](https://adk.dev/tools-custom/confirmation/) - [Claude: yes — Claude is supported in Python and Java, either through the registry wrapper on Google Cloud Agent Platform or directly against Anthropic with your own API key.](https://adk.dev/agents/models/anthropic/) **Trade-offs against AutoGen:** - [Writes code: partial — ADK does not write application code for you, but its Visual Builder inside the local web UI generates agent projects from a graphical design with an AI assistant, and the vendor marks that builder experimental and Python-only.](https://adk.dev/visual-builder/) - [In a browser: partial — The browser UI exists but you run it locally and the vendor bars it from production, so there is no hosted web application unless you deploy your own agent service.](https://adk.dev/runtime/web-interface/) [View Google Agent Development Kit (ADK)](https://theagentsindex.com/google-adk) ### 4. Pydantic AI Open-source Python agent framework from the Pydantic team, with a typed agent loop and swappable models. **Research coverage:** 34/35 criteria **Where it improves on AutoGen:** - [Agent permissions: yes — Tools can be marked as requiring human approval, and the run pauses with the pending calls so a person approves or denies each one before the tool body executes; the docs also warn approval is not an authorization boundary.](https://pydantic.dev/docs/ai/tools-toolsets/deferred-tools/) - [Claude: yes — Anthropic is a first-class built-in provider with its own model class and install extra, selectable by a provider-prefixed model string like anthropic:claude-sonnet-4-6.](https://pydantic.dev/docs/ai/models/overview/) - [Open models: yes — Open-weight models are reachable through OpenAI-compatible providers the docs enumerate, including local runtimes Ollama and vLLM as well as Hugging Face and Groq, so weights you host yourself are supported.](https://pydantic.dev/docs/ai/models/overview/) **Trade-offs against AutoGen:** - [Writes code: partial — The framework itself is a Python SDK, but its Harness ships a Coder capability whose file tools let an agent author and edit files in a workspace, so code writing comes from the bundled agent rather than a product you sign into.](https://pydantic.dev/docs/ai/harness/) - [Getting out: partial — Prepaid Gateway credits are non-refundable and expire after twelve months of account dormancy; Logfire data is exportable over its query API and Growth adds self-serve deletion.](https://pydantic.dev/legal/terms-of-service) [View Pydantic AI](https://theagentsindex.com/pydantic-ai) ### 5. Letta Stateful agent harness, cloud and SDK for agents with persistent, git-versioned memory. **Research coverage:** 35/35 criteria **Where it improves on AutoGen:** - [Debugs and fixes: yes — Vendor states coding agents debug failures and run test suites.](https://docs.letta.com/handbook/index.md) - [Reviews code: yes — Reviews diffs on demand or on every PR via the GitHub action's prompt input.](https://docs.letta.com/platform/github-action/index.md) - [Agent permissions: yes — Four modes from strict to unrestricted, plus allow/deny tool patterns and a cross-agent memory guard.](https://docs.letta.com/configuration/permissions/index.md) **Trade-offs against AutoGen:** - [MCP server: no — MCP runs inbound only: Letta sessions consume stdio, HTTP and SSE MCP servers; no Letta MCP server is published.](https://docs.letta.com/agent-sdk/mcp/index.md) - [Audit log: no — No administrative audit log is published; the Enterprise list stops at RBAC, SSO, quotas and dedicated support.](https://docs.letta.com/pricing/index.md) - [Getting out: partial — Memory is git-backed files you can inspect and sync to your own GitHub repo; credit expiry is never stated.](https://docs.letta.com/handbook/index.md) [View Letta](https://theagentsindex.com/letta) ### 6. CrewAI Open-source multi-agent framework with a managed platform for deploying and governing agent workflows. **Research coverage:** 34/35 criteria **Where it improves on AutoGen:** - [Done-for-you service: partial — You run the software yourself, but the Enterprise plan sells forward-deployed engineers, onboarding and training separately, so people-delivered build help exists as a paid extra.](https://www.crewai.com/pricing) - [Claude: yes — Anthropic is one of the natively integrated providers, configured with your own key, and Claude models are also reachable through Bedrock and Snowflake Cortex.](https://docs.crewai.com/en/concepts/llms) - [Open models: yes — Local and open-weight serving is documented end to end, with step-by-step Ollama setup plus Hugging Face, Together, Groq and local NVIDIA NIM examples via LiteLLM.](https://docs.crewai.com/en/learn/llm-connections) **Trade-offs against AutoGen:** - [Writes code: partial — Agents you build can author and run Python when code execution is switched on, but this is a capability of agents inside the framework rather than a product that ships code for you.](https://docs.crewai.com/en/learn/coding-agents) - [Getting out: partial — Automation source can be pulled back as a ZIP over the API and traces can be shipped to your own OpenTelemetry collector; nothing published covers rollover of unused executions.](https://docs-platform.crewai.com/api/v1/reference/automations/export-an-automation-zip) [View CrewAI](https://theagentsindex.com/crewai) ## Method Unknown data is never treated as a benefit or a trade-off. Featured placement is labelled but cannot alter recommendation order.