Agent Settings Reference
This page lists every setting on an agent. Open an agent from Build → Agents and you'll find the fields on its Configure tab. For what agents are and how the settings fit together, see Agents.
Configure tab: fields
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
| Agent Name | Text | Yes | — | Shown in chat, teams, workflows and logs. Must be unique within the project. Other agents use this exact name to call it as a sub-agent or hand a chat to it. |
| System Message | Multiline text | Yes | — | The agent's instructions: its role, scope, output format and constraints. |
| Greeting | Multiline text | No | Empty | What the agent says first, before the user types anything, in a one-to-one chat. Leave it empty and the chat opens on a blank screen. |
| Opening suggestions | Multiline text, one per line | No | Empty | One-tap prompt chips shown beside the greeting until the user types something of their own. Blank lines are ignored. |
| LLM Configuration | Select | Yes | — | The model that powers the agent. Platform-provided models carry a Platform chip. If you don't choose one, saving fails with "Invalid LLM Config." See Models. |
| Tags | Text, comma-separated | No | Empty | Labels for organising and searching. They don't change behaviour. |
| Default Role | Text | No | Empty | A short description of what the agent is for. Teams use it to describe the agent to the other members, and a Group Chat Selector coordinator reads it to choose who speaks next. A team-level Role for the agent replaces it. If neither is set, the agent is described as "An agent that provides assistance". |
| Approval Override | Select | No | Inherit from risk appetite | Inherit from risk appetite leaves the risk-appetite tier to decide. Always require (pin) makes every tool call wait for a person. Never require (pin) skips approval, except where critical-risk approval still applies. See How approval is decided. |
| Execution Environment | Select | No | Project Default | Where the agent's tools run. Project Default uses the project's default environment. See Execution Environments. |
| Memory Policy | Select | No | Platform Default | How much the agent remembers across sessions. Takes effect only when Memory is ticked. Options are listed below. |
Memory Policy options
| Option | Behaviour |
|---|---|
| Platform Default | Uses the platform-wide policy, which is Knowledge Worker. |
| Off | Stores and injects nothing. |
| Conversational (short retention) | Keeps episodic memories for 14 days, with small caps and a small injection budget. |
| Knowledge Worker (durable) | Keeps episodic memories for 90 days, including learned procedures, with a larger injection budget. |
| Compliance (curated) | Keeps episodic memories for 3,650 days, with no row caps and no decay. |
See Memory for how the tiers work.
Configure tab: other controls
| Control | Description |
|---|---|
| Profile Picture | Click the avatar to upload an image (JPEG, PNG or WebP, up to 2 MB). It's used in chat and lists. This control appears only once the agent has been created. |
| Assigned Functions | The tools the agent can call, shown as chips. Click a chip to open the function. Click its delete icon to unassign it. |
| Assign Functions | Opens Assign Functions to Agent. Search, multi-select, then click Assign. Platform-provided functions carry a platform badge. |
| Silent | Checkbox, off by default. Stored on the agent and copied with it, but it has no visible effect on chat behaviour today. |
| Streaming | Checkbox, off by default. Streams the agent's replies token by token in a one-to-one chat. Inside a team, the team's Enable Streaming setting applies instead. |
| Memory | Checkbox, off by default. Turns on long-term memory. The agent gets relevant memories injected at the start of each chat, plus the remember and recall built-in tools. Memories are distilled when sessions end. |
Tabs
| Tab | What it holds |
|---|---|
| Overview | Summary cards: System prompt, Risk and compliance, Readiness, At a glance, Tools, Reach, Activity · 30 days, Tests, Used by and Effective policy. |
| Configure | The fields and controls above. |
| Topology | A map of what the agent can reach (tools, integrations, trust boundaries, external hosts, model), with a Flow or Threat model layout and findings. |
| Tests | Evaluation test cases. See Testing Agents and Teams. |
| Sub-agents | The agents this agent may invoke or hand a chat to. Pick one under Agent, click Grant, and remove a chip to revoke it. |
| Repositories | Repositories this agent works on: GitHub repositories the agent is told about at the start of every chat. This is context, and it limits the agent's calls made through the FYAI GitHub App to these repositories. See GitHub repositories. |
| Memory | Attached knowledge bases, memory readiness, Memory sharing, and the agent's stored memories (search, edit, delete, Consolidate now, Audit log, Clear all). See Memory. |
| Observability | Effectiveness Metrics, Chat Efficiency, Recent Errors and Audit Trail. See Observability. |
| Risk | Risk Level, Data Classification, Risk Owner, Risk Categories, Compliance Tags, Risk Notes, Require justification before starting a session, Mark as Reviewed. |
| Controls | Governance control evidence for this agent. See Governance. |
| Identity | The agent's platform identity and its Tool Identity overrides. See below. |
| History | Every configuration revision, with diffs and Restore. See Observability. |
Identity tab
| Section | Contents |
|---|---|
| Identity overview | Shows the identity provider, Active or Disabled, Provisioned or Not provisioned, and the client type. |
| Credentials & Endpoints | Read-only: Client ID, Application Slug, Provider Type, Client Type, Signing Key, Token URL, Access Token Validity, Refresh Token Validity, Issuer Mode, Sub Mode, Created, Provisioned By. |
| Security Health | Three checks: the custom-claims scope is attached, the audience scope is attached, and the identity is active. |
| Danger Zone | Disable Identity or Enable Identity. While the identity is disabled the agent can't obtain new tokens; tokens it already holds stay valid for up to 10 minutes. Currently only platform administrators can change this; for anyone else the change fails with "Failed to update identity status." |
| Tool Identity | Overrides that redirect which integration a tool uses for this agent. Each row is When a tool asks for (the integration name in the tool's code) and Act as (one of your integrations). Use Add override, Update and Reset. An override can only point at an integration the agent is already allowed to use. |
Header and actions
| Control | Description |
|---|---|
| Active / Killed switch | The killswitch. Turning it off opens Kill this agent? and offers Also terminate active chat sessions for this agent. A killed agent can't run anywhere, and its configuration is read-only until you reactivate it. |
| Open in Chat | Starts a new one-to-one conversation with the agent in the docked chat. |
| Update | Saves the Configure fields. |
| Cancel | Discards unsaved changes. |
| Copy Agent | Opens Copy Agent. Enter a New Agent Name (default: the current name followed by "(copy)") and click Copy. The copy gets the system message, model, tags, Silent / Streaming / Memory, default role, approval override, execution environment and functions; it does not get the greeting, opening suggestions, memory policy, sub-agent grants, repositories, tool identity or risk fields. |
| Publish Template | Platform administrators only. Publishes the agent as a platform template. See Templates. |
| Delete | Asks "Delete this agent?" and then deletes the agent. Requires the project Admin role. |
Platform-managed agents are labelled Managed by platform — read only, and non-administrators can't edit or delete them.
Fields available only through the API
These fields exist on the agent record but aren't in the editor. You can set them with the Platform API (PATCH /agents/{id}) or the Platform MCP server.
| Field | Description |
|---|---|
description | Free-text description. |
path | The folder path in the Agent Library. In the UI you set it by dragging the agent into a folder. |
integration_ids | Integrations the agent is granted access to. |
integration_overrides | The Tool Identity map. The UI edits this on the Identity tab. |
mcp_tool_overrides | Per-agent hiding of individual MCP server tools. |
repository_ids | The Repositories tab list. |
agent_type | Free-text label. It's stored and copied, but it has no effect on how the agent runs. |
risk_level, data_classification, risk_categories, compliance_tags, risk_notes, risk_owner, require_justification | The Risk tab fields. |
is_active | The killswitch state. Change it with the killswitch (POST /killswitch/agent/{id}) rather than PATCH. |
Related API endpoints
| Method and path | Purpose |
|---|---|
GET /agents, POST /agents | List agents, or create one. Creating requires the Editor role. |
GET /agents/search?field=&query=&fuzzy= | Search agents by a field. |
GET /agents/{id}, PATCH /agents/{id}, DELETE /agents/{id} | Read, update or delete an agent. Updating requires Editor and deleting requires Admin. |
POST /agents/{id}/functions, DELETE /agents/{id}/functions/{function_id} | Assign or remove functions. |
GET / POST /agents/{id}/sub-agents, DELETE /agents/{id}/sub-agents/{target_id} | Sub-agent grants. |
POST /agents/{id}/profile-picture | Upload an avatar. |
GET /agents/{id}/readiness | Check whether the agent can run, without calling the model. |
POST /agents/{id}/run | Run the agent once and wait for the reply. |
POST /agents/{id}/run-async, GET /agents/run-async/{execution_id} | Start a run in the background, then poll for its result. |
GET /agents/{id}/identity, PATCH /agents/{id}/identity/active | Read the agent's identity, or enable and disable it. |
POST /agents/{id}/risk-review | Record a risk review. |