Agent Frameworks
Frameworks and tools for building AI agents and multi-agent workflows.
43 results in Agent Frameworks
Active Pro listings from this category appear first. Your sort applies within Pro and Free results.
Shortlists for Agent Frameworks
Explore collections with tools relevant to this category, grouped by deployment, source model or workload.
Self-hosted IT tools →
38 matching tools from this category
IT tools with an API →
35 matching tools from this category
Open-source IT tools →
39 matching tools from this category
How to choose agent frameworks tools
Agent frameworks give developers the building blocks for software that plans, calls tools and acts on behalf of a user: model abstraction, tool definitions, memory, retrieval and orchestration of several cooperating agents. Many also ship a runtime that executes agents as long-running services with task queues, approvals and cost tracking. They sit below finished assistants and coding agents, since you assemble the behaviour yourself rather than install a ready product.
Who uses these tools
- Developers building agentic features into products
- Platform teams running internal agent services
- Machine learning engineers designing retrieval pipelines
Capabilities to compare
- Tool calling and function definition interfaces
- Short and long term agent memory
- Multi-agent orchestration and task handoff
- Retrieval over documents and knowledge bases
- Tracing, evaluation and human approval steps
Operating considerations
- Isolation of agents that run shell commands or browsers
- Token spend multiplies quickly with multi-agent loops
- Abstractions can tie code to one framework's conventions
- Hosted runtimes may store prompts, traces and tool outputs
Selection checklist
- Build one small agent end to end with your real tools before committing to a framework.
- Check which model providers and local inference servers the framework supports without custom adapters.
- Inspect how runs are traced, so you can replay a failure and see each tool call.
- Confirm how permissions, approvals and sandboxing work before giving any agent write access to production systems.
What belongs in this category: A product fits here if its main purpose is to let developers define, run or coordinate their own agents through code, configuration or workflow files, rather than offering a finished chat or coding assistant.
Agent Frameworks on iThub in figures
iThub lists 43 tools in Agent Frameworks. 39 of 43 are open source, 38 of 43 can be self-hosted, 42 of 43 have a free, freemium or open-source pricing model and 35 of 43 offer an API.
- Tools listed
- 43
- Open source
- 39
- Self-hostable
- 38
- Free to start
- 42
- With an API
- 35
- Web app
- 28
- Desktop app
- 30
Recently added: OpenComputer, Omnara, Sugabots, Dify
Typical use cases
- Answering internal questions from company documents with retrieval
- Automating multi-step support or operations tasks with approvals
- Coordinating several specialised agents on a research workflow
- Running scheduled agents that report on systems or data
Trade-offs to model
- A thin library stays close to the model API and is easy to debug, while a full platform adds runtime features and more lock-in.
- Self-hosting the agent runtime keeps prompts and data inside your network, whereas managed runtimes remove operational work but add a data processor.
Common mistakes
- Starting with a multi-agent design when a single agent with good tools would solve the problem more cheaply.
- Giving agents broad credentials during prototyping and never narrowing them before the workflow reaches real users.
- Skipping evaluation sets, so nobody notices when a prompt or model change quietly degrades results.
Compare LangChain and LlamaIndex →Explore MCP servers →See MLOps and LLMOps tools →
Continue the technical evaluation
Use the resources that match your decision stage. Profile data and linked sources were refreshed through November 11, 2026.
