Agentfield
Build, run and scale AI agents like API and microservices - observable,auditable and identity-aware from day one.
Last verified:
What is Agentfield?
AgentField is an open-source AI backend platform that enables developers to build, deploy, and govern AI agents the way they build APIs. It serves as a control plane for autonomous software, providing infrastructure for agents to operate with cryptographic identity, enforceable permissions, and tamper-proof audit trails. The platform addresses the coordination crisis in autonomous software by merging Kubernetes-like orchestration with decentralized identity features, allowing agents to negotiate, act, and complete intent within well-defined boundaries.
Key features include: APIs where decorated functions become HTTP endpoints with discovery and tracing; support for 100+ LLMs with structured output (Pydantic/Zod/structs) and tool calling; multi-agent capabilities with app.call, shared memory, async execution, webhooks, and governance (DIDs, policy, audit); human-in-the-loop approval workflows; signed webhooks; cross-agent discovery; and SDKs for Python, Go, and TypeScript. The platform scales to thousands of agents with no timeout limits and supports canary deploys, A/B testing, and blue-green rollouts.
AgentField is designed for developers building AI agents, engineering teams deploying autonomous software to production, companies needing governance and audit trails for AI actions, and organizations handling financial transactions or sensitive data where trust and verification matter. It's particularly valuable for insurance claims processing, enterprise automation, and any use case where agents touch money, data, and decisions.
Agentfield pricing
Pricing model: Freemium
AgentField is open-source and free under Apache 2.0 license, available on GitHub. The software itself has no cost - you pay only for your own LLM API calls (BYO-API) or can use free local models via Ollama. Infrastructure costs depend on your self-hosting setup (estimated $5-50/month for light setups on VPS or hardware). There are no paid tiers from AgentField since it's a self-hostable control plane.
Agentfield pros
- Open-source under Apache 2.0 license
- Cryptographic identity (W3C DIDs) for each agent
- Tamper-proof audit trails for all agent actions
- Kubernetes-like orchestration for AI agents
- 100+ LLMs supported out of the box
- Structured output with Pydantic/Zod/structs
- Multi-agent support with app.call and shared memory
- Async execution with no timeout limits
- Signed webhooks for secure communication
- Cross-agent discovery capabilities
- Human-in-the-loop approval workflows
- Python, Go, and TypeScript SDKs available
- Scales to thousands of agents
- One-line production deployment with docker compose
- Canary deploys, A/B testing, blue-green rollouts
Agentfield cons
- Open-source means self-hosting required
- No managed cloud service offered
- Requires Docker knowledge for deployment
- Steep learning curve for governance features
- Best suited for production use cases
- Limited documentation for beginners
- Requires BYO API keys for LLMs
- New platform with smaller community
Frequently asked questions about Agentfield
What is AgentField?
AgentField is an open-source AI backend platform and control plane that lets you build, deploy, and govern AI agents like APIs. It provides Kubernetes-like orchestration with cryptographic identity (W3C DIDs), enforceable permissions, and tamper-proof audit trails for autonomous software.
Is AgentField free?
Yes, AgentField is completely free and open-source under the Apache 2.0 license. The software costs nothing - you only pay for your own LLM API calls or can use free local models. Infrastructure hosting costs depend on your self-hosting setup.
How do I install AgentField?
Install with one command: curl -sSf https://agentfield.ai/get | sh. After install, type /agentfield in Claude Code or describe your system in plain English. The full stack becomes live under docker compose in one shot.
What programming languages does AgentField support?
AgentField provides SDKs for Python, Go, and TypeScript. The example code shows Python with Pydantic for structured output, but you can build agents in any of these three languages.
How does AgentField handle agent identity?
Each agent is assigned a W3C Decentralized Identifier (DID) providing cryptographic identity. This ensures when AI performs actions like financial transactions or accessing sensitive data, it does so under enforceable permissions with a mathematical basis for authority.
What LLMs are supported?
AgentField supports 100+ LLMs with structured output capabilities. You can use models from Anthropic (like claude-sonnet-4), OpenAI, and others. You bring your own API keys or use free local models via Ollama.
Can AgentField handle multi-agent workflows?
Yes, AgentField supports multi-agent orchestration with app.call for cross-agent calls, shared memory between agents, async execution, webhooks, and governance. Agents can discover each other and route execution through the control plane with full tracing.
What is human-in-the-loop in AgentField?
Human-in-the-loop allows agents to pause for human approval when needed. For example, if confidence is below 0.85, an agent can call app.pause() with an approval_request_id, suspend execution, notify humans, and resume when approved within a set expiration time.
How does AgentField compare to frameworks?
AgentField is not just an app feature but an AI backend infrastructure. Unlike frameworks, it provides HTTP endpoints, routing, memory, tracing, policy, and async execution as infrastructure. It acts as the control plane for autonomous agents like Kubernetes serves containers.
Who founded AgentField?
AgentField was founded by Oktay Goktas, PhD and Santosh Kumar Radha, PhD, the team behind Agnostiq (acquired by DataRobot in 2025) and Covalent. The company continues a decade of work on distributed computing and infrastructure for high-stakes systems, headquartered in Toronto.