Agent Manage Tool
agent_manageManage AI agents: create, list, update, delete, and toggle status. Browse pre-built agent templates.
Instructions
Core CRUD for AI agents in the caller's team. For runtime inspection, rollback, skill/tool wiring or feedback use agent_advanced. provider and model are validated against team-configured BYOK and local-LLM credentials at create/update time.
Actions:
list (read) — optional: status, limit (default 50), cursor.
get (read) — agent_id.
create (write) — name, role, goal; optional: backstory, provider, model, skill_ids[], tool_ids[].
update (write) — agent_id + any creatable field. Partial updates allowed.
delete (DESTRUCTIVE) — agent_id, confirm=true. Soft-deletes; recoverable for 30 days.
toggle_status (write) — agent_id. Flips active ↔ disabled.
templates (read) — pre-built agent templates from the platform catalog.
Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| action | Yes | Action to perform: list, get, create, update, delete, toggle_status, templates | |
| deadline_ms | No | Optional: max wall-clock time (ms) the tool may spend. If exceeded during the call, returns a DEADLINE_EXCEEDED error. Minimum 100 ms. Leave unset for no deadline. | |
| status | No | Filter by status: active, disabled | |
| scope | No | Filter by scope: team, personal | |
| limit | No | Max results to return (default 10, max 100) | |
| agent_id | Yes | The agent UUID | |
| name | Yes | Agent name | |
| role | No | Agent role description | |
| goal | No | Agent goal | |
| backstory | No | Agent backstory | |
| provider | No | LLM provider key (e.g. anthropic, openai, google, claude-code). Defaults to platform default. | |
| model | No | LLM model name. Defaults to platform default. | |
| personality | No | Agent personality traits: {tone, communication_style, traits[], behavioral_rules[], response_format_preference} | |
| data_classification | No | Data classification level: public, internal, confidential, restricted. Confidential and restricted agents are routed to local-only providers. | |
| tool_profile | No | Tool profile restricting tool access. Options: researcher, executor, communicator, analyst, admin, minimal | |
| environment | No | Environment preset that auto-attaches a tool bundle. Options: minimal, coding, browsing, restricted. | |
| reasoning_effort | No | Extended thinking effort (Anthropic). Options: none, low, medium, high, auto. "auto" lets the platform pick by task complexity. | |
| use_tool_search | No | Enable semantic tool auto-discovery — at run time, up to tool_search_top_k matching tools from the team pool are auto-attached based on the user prompt. | |
| tool_search_top_k | No | Maximum tools tool_search will surface per run (1–20, default 5). Only applies when use_tool_search=true. | |
| sandbox_profile | No | JSON string defining Docker sandbox profile for per-execution process isolation (enterprise only). Example: {"image":"python:3.12-alpine","memory":"512m","cpus":"1.0","network":"none","timeout":300} | |
| knowledge_base_id | No | UUID of a knowledge base to link to this agent for RAG-powered context | |
| evaluation_enabled | No | Enable A/B evaluation for this agent | |
| evaluation_sample_rate | No | Fraction of requests to include in evaluation (0.0 to 1.0). Only used when evaluation_enabled is true. | |
| heartbeat_definition | No | Agent health check config: {enabled: bool, cron: string, prompt: string} | |
| budget_cap_credits | No | Per-agent budget cap in credits. Set to 0 to remove cap. | |
| thinking_budget | No | Anthropic extended thinking budget in tokens (e.g. 1024, 4096, 8192). Only applies when agent provider is "anthropic". Set to 0 to disable. Enables chain-of-thought reasoning visible in experiment steps. | |
| confirm | Yes | Must be true to confirm deletion. This is a destructive action. | |
| category | No | Filter by category: engineering, content, business, design, research |