What gets recorded
- Each agent's tool calls, sealed by an adapter inside the framework you already use (CrewAI, LangChain, MCP and others).
- For A2A, a task-evidence record that travels with the task between agents.
Use cases · Multi-agent & A2A task evidence
When a task passes between agents, each hand-off leaves a record.
In a multi-agent system, one agent plans, another calls tools and a third reports back. When something goes wrong, each framework's own trace describes its own part, and none of them can show a stranger which agent did what.
What anyone can check
- Every record verifies offline with the open verifier, whichever framework produced it.
What runs today
What's built, and what isn't yet.
- Runs today Framework adapters in capsule-emit ↗ (see integrations).
- Preview A2A task evidence: an example ↗ and a draft A2A extension ↗.