Jami Agent Harness
Run agents through one governed loop where tools, memory, approvals, model routing, and user-facing actions share the same contract.
Who it serves
Developers building agent-native products that need control, auditability, and provider choice.
Agent-readable shape
Harness is the Studio runtime layer for governed agent loops, durable runs, provider abstraction, tool contracts, memory, and policy-gated dual invocation.
What it provides
- Governed run loop with policy checks before every action
- BYOK-friendly engine registry and model catalog
- Shared contracts for human UI actions and agent tool calls
- Durable run state, memory, and resumable execution
Why the boundary holds
- Designed around one enforcement path instead of UI backdoors
- Provider choices live in adapters and configuration
- Keeps runtime implementation outside the marketing site