what are the top 10 agentic frameworks?

the ten, grouped by what they are for

Graph-based control flow. You draw the state machine and the agent moves through it.

• Langgraph. Explicit nodes, edges, and state, with checkpointing and human-in-the-loop pauses. The most control, the steepest start.
• Llamaindex workflows. Event-driven steps, strongest when retrieval is the centre of the system.

Role-based teams. You describe who does what and the framework handles the sequencing.

• Crewai. Agents with roles and goals, tasks with expected outputs, sequential or hierarchical process. Fastest route to a working team.
• Autogen. Microsoft research's conversational multi-agent framework, group chat patterns and termination conditions.

Provider sdks. Thin layers from the labs themselves.

• Openai agents sdk. Agents, handoffs, guardrails, and tracing. The successor to Swarm.
• Claude agent sdk. The harness behind claude code, exposed for your own agents.

Typed application frameworks. Ordinary software engineering with agents in it.

• Pydanticai. Type-safe, validated structured output, familiar to anyone who uses pydantic.
• Mastra. Typescript, batteries included, workflows and memory in one place.
• Semantic kernel. Microsoft's enterprise-facing sdk across c#, python, and java.
• Smolagents. Hugging face's minimal one, where agents write python to call tools. Small enough to read in an evening.

how to choose in two minutes

• Need explicit control of state and resumable runs: langgraph.
• Need a team of roles working today: crewai.
• Already committed to one provider: that provider's sdk.
• Typescript codebase: mastra.
• Want to read the whole framework before trusting it: smolagents.
• Enterprise dotnet or java: semantic kernel.

what none of them give you

This is the part that decides how long the project takes. No framework in the list ships permissions, tenant isolation, durable memory with pruning, spend caps, an audit trail you can search, or a human gate before irreversible actions. Every team builds those, identically, and it is roughly a quarter of engineering rather than a weekend.

The framework is genuinely the easy part. The loop is about eighty lines. Everything around it is the project.

how this works in aldena

If the agent is your product, pick a framework from the list and build. If it is a means to an end, aldena is the layer you would otherwise write on top of one.

Agents come as roles, eleven prebuilt, arranged in an org chart rather than wired with handoff code. The list of missing pieces above is the product: isolated rooms with their own server, durable memory at 50 private and 100 shared entries, per-agent tool permissions, and an approval gate on by default. Model choice stays yours per agent, from the published catalog.