Skip to main content
Hyperterse agents are declarative, ADK-backed runtimes discovered from your project filesystem, just like tools. You define agent behavior in config files; Hyperterse compiles, validates, and mounts each agent behind a dedicated HTTP route prefix.

​
What you get

  • Declarative agent discovery from app/agents/**/config.terse
  • Compile-time permission checks for tool access policies
  • Per-agent runtime endpoints for run, streaming, and sessions
  • Multi-provider model config for Gemini, Vertex AI, and OpenAI-compatible backends

​
How the system fits together

.hyperterse
app/
  tools/
    get-orders/config.terse
  agents/
    support/config.terse
  1. Compiler discovers tools and agents.
  2. Tool access policy is resolved (inherit / allow_*).
  3. Agent definitions are compiled into runtime model.
  4. Runtime mounts ADK endpoints under /agent/{agentName}.

​
Endpoint shape per agent

Every compiled agent gets:
  • /agent/{agentName}/run
  • /agent/{agentName}/run_sse
  • /agent/{agentName}/list-apps
  • /agent/{agentName}/apps/{app_name}/users/{user_id}/sessions...
Agent endpoints are separate from MCP /mcp. Use MCP for tool-native workflows, and use /agent/* when you need ADK-style conversational/session behavior.

​
Reading path

Use this order for fastest onboarding:
  1. Quickstart — create and run your first agent.
  2. Tool access — lock down permissions correctly.
  3. Runtime API — understand request/response + sessions.
  4. Model providers — configure provider-specific models.

​
References