// the take
uAgents is Fetch.ai's Python framework for building autonomous agents that communicate over a custom messaging layer, with each agent registering itself on the Almanac smart contract on Fetch.ai's blockchain. It's aimed at developers who want agent-to-agent communication and discovery baked in, not just a single-process LLM wrapper, and includes adapters for LangChain, CrewAI, and MCP to bridge into other agent stacks.
The decorator-based API (on_interval, on_message, on_event) is genuinely minimal and gets a working agent running in a handful of lines. Message signing/verification and wallet handling are built in rather than bolted on, so agent identity and message integrity come for free. The adapter packages (uagents-adapter for LangChain/CrewAI/A2A/MCP) are a real differentiator if you need uAgents-style discovery on top of an existing agent built elsewhere. CI is active with real test coverage (test_registration, test_resolver, test_msg_verify) rather than just smoke tests.
Every agent registers on the Almanac blockchain contract on startup, which means your 'lightweight' agent now has a hard dependency on Fetch.ai's chain being up and reachable — that's an unusual amount of infrastructure coupling for what's pitched as a simple framework. A lot of the actually useful functionality (dialogues, mobility, chat_agent, quota, search) lives under experimental/ with no stability guarantees, so you're building on APIs that can change under you. Observability is tied to Agentverse, Fetch.ai's hosted dashboard, rather than standard logging/tracing you could point at your own stack. The mailbox/resolver model for agent-to-agent discovery is a proprietary addressing scheme, not a protocol like A2A or MCP natively, so interop outside the Fetch.ai ecosystem relies entirely on the adapter packages staying maintained.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →