Agentspan

Open-source runtime adding durability and safety to any AI agent.

Agentspan is an open-source runtime that adds durability features — crash recovery, human-in-the-loop approvals, guardrails, tool history, and observability — to any existing agent framework or LLM, rather than requiring teams to adopt an entirely new agent framework. Many teams build agents using their preferred framework or directly against an LLM API, but lack the production-readiness features needed for reliable operation, such as recovering gracefully from crashes or maintaining an audit trail of agent actions. Agentspan addresses this by providing a runtime layer that can wrap around existing agent implementations, adding these capabilities without requiring a framework migration. It is available under the MIT license and supports both self-hosted and cloud deployment, targeting SMB and mid-market development teams that have already built agents and need to harden them for production use.

Compliance

SOC 2

Visit Agentspan

Key Features

Crash recovery: Provides automatic recovery from crashes during agent execution, preserving state and allowing agents to resume rather than losing progress on long-running tasks.
Human-in-the-loop approvals: Adds configurable approval checkpoints to agent workflows, allowing teams to require human sign-off on specific actions regardless of the underlying agent framework.
Guardrails: Provides a guardrail layer that can constrain agent behavior and actions, adding a safety control that may not be natively present in the underlying agent framework.
Tool history and audit trail: Maintains a record of tool calls and actions taken by the agent, supporting debugging, auditability, and compliance needs.
Observability: Adds monitoring and observability capabilities to agent execution, giving teams visibility into agent behavior without building custom instrumentation.
Framework-agnostic design: Works with any existing agent framework or LLM rather than requiring adoption of a new framework, making it additive to a team’s current agent implementation.

Use Cases

For teams hardening existing agent implementations for production: A team that has built an agent using their preferred framework but lacks production-readiness features uses Agentspan to add crash recovery, observability, and guardrails without rebuilding their agent from scratch.
For teams needing audit trails for agent actions: An organization that needs to demonstrate what actions its AI agents have taken, for compliance or debugging purposes, uses Agentspan’s tool history feature to maintain a structured audit trail.

Pricing MODELS

Open-source free

Pricing Summary

The Agentspan runtime is free and open source under the MIT license. Paid cloud plans are available for managed hosting; specific cloud pricing is not publicly detailed.