Execute ad-hoc agent
Creates an ephemeral agent and immediately executes a run. Returns the run ID for polling status and results. This is the fastest way to run an agent without managing agent lifecycle.
Ephemeral agent lifecycle: The agent created by this endpoint is automatically soft-deleted and its scoped API key revoked when the run completes (whether it succeeds, fails, or times out). Ephemeral agents do not appear in GET /agent/v1/agents listings. The returned agentId is valid only for the duration of the run — do not store it for reuse. For persistent, reusable agents, use POST /agent/v1/agents instead.
Authorizations
API key starting with sk_case_
Body
Task prompt for the agent
System instructions. Defaults to a general-purpose legal assistant prompt if not provided.
LLM model identifier. Defaults to anthropic/claude-sonnet-4.6
Restrict agent to specific vault IDs
Scope this run to specific vault object IDs. The agent will only access these objects.
Additional context or constraints for this run
Allowlist of tools the agent can use. Mutually exclusive with disabledTools — set one or the other, not both.
Denylist of tools the agent cannot use. Mutually exclusive with enabledTools — set one or the other, not both.
Custom sandbox resources (cpu, memoryMiB)