Skip to content

Governance Policies

5. Agent Management

Agent Management is the tenant-level policy configuration centre that governs what capabilities all agents in the tenant are permitted to use. Unlike the per-agent settings in Agent DNA, this page defines the outer boundaries that apply to every agent regardless of their individual configuration.

5.1 Policy Summary Stats Bar

A live metrics bar displays the current effective policy state:

Metric Description
Provider Mode Whether the tenant is in permissive, whitelist, or blacklist mode for LLM providers
Active Agents Total number of currently active agents across the tenant
Max Keys / Agent The maximum number of API keys any single agent may hold
Key Expiry Whether API key expiry is required or optional
Max Agents The maximum number of agents allowed for this tenant (Unlimited if 0)
Guard Whether all agents are required to have a guard assigned

5.2 Provider Controls

Configures which LLM providers and models agents may use:

Provider Mode

  • Permissive — All providers are allowed unless explicitly blocked
  • Whitelist — Only listed providers are allowed
  • Blacklist — All providers are allowed except those explicitly blocked

Allowed / Blocked Providers

Tag-based input fields to add or remove specific provider names (e.g., openai, anthropic, azure).

Per-Provider Model Controls

Fine-grained control to allow or block specific models within a provider (e.g., allow gpt-4o but block gpt-4-turbo).

Fallback Providers

Configures up to three automatic fallback providers and models. If the primary provider is unavailable or blocked, the system attempts the fallback chain in order. Each fallback can specify allowed models.

5.3 Feature Controls

Controls which model and agent capabilities are available tenant-wide:

Model Capabilities

  • Text — Always enabled; base text generation
  • Vision — Image input/output
  • Audio — Audio input/output
  • Video — Video input/output
  • Tool Calling — Function/tool execution (required for most agent capabilities)

Agent Capabilities

The following require Tool Calling to be enabled: - MCP — Model Context Protocol connections - RAG / Retrieval — Document retrieval augmented generation - Web Search — Internet search access - Runtime Policy — Permission to modify runtime policy - Tool Scope — Permission to manage tool scope - MCP Scope — Permission to manage MCP scope

5.4 Key & Deletion Management

API Key Settings

  • Require Key Expiry — Toggle that forces all agent API keys to have an expiration date
  • Default Key Expiry (hours) — The default expiration period when keys are created
  • Max API Keys per Agent — Maximum keys a single agent may hold simultaneously

API Key Deletion Policy

Defines what happens when a key is revoked: - Immediate — Key is deleted on revocation - Time-based — Key is deleted after a configured grace period (hours) - Traffic-based — Key is deleted after a configured period of inactivity (hours)

Agent Deletion Policy

Defines what happens when an agent is removed: - Immediate — Agent is deleted on request - Time-based — Agent enters a grace period before deletion

When a restrictive change is applied (e.g., reducing max keys or blocking a provider), the system calculates the number of affected agents and grace period hours, and displays a notification toast.

5.5 Resource Defaults

Configures the default tool, MCP, skill, and RAG collection access for newly deployed agents:

  • Default Tools Mode — permissive (all tools allowed) or allowlist / blocklist
  • Default Allowed / Blocked Tools — Tag lists
  • Default MCPs Mode — permissive or allowlist / blocklist
  • Default Allowed / Blocked MCPs — Tag lists
  • Default Skills Mode — permissive or allowlist / blocklist
  • Default RAG Mode — permissive or allowlist / blocklist

5.6 Guard Department Controls

  • Required Guard — Toggle that forces all agents to have a governance guard assigned
  • Allowed Guard Domains — Comma-separated list of guard domains that agents may use (empty = all domains allowed)
  • Guard Department Mode — allowlist (only listed departments may use guards) or permissive
  • Guard Allowed Departments — Dropdown checklist of departments from the employee registry

If configured allowed guard domains conflict with any currently deployed guards, a warning banner lists the conflicting guard names.

5.7 Change Reason & Restrictive Changes

When a change to the policy is classified as restrictive (e.g., reducing key limits, switching from permissive to whitelist mode, blocking a previously allowed provider), operators must provide a written Change Reason of at least 15 characters before saving.

Non-restrictive changes can be saved without a change reason.

5.8 Change History

The Change History button opens a paginated timeline of all past policy changes, including: - Change type (create / update / reset) - Timestamp - Change reason (if provided) - List of changed fields - Grace period trigger and end date (if applicable) - Before/after diff expandable section

6. Policy Compliance

Policy Compliance tracks whether each deployed agent is compliant with the current tenant policy. It surfaces agents that are violating policy, in a grace period, or suspended — and provides drill-down into the specific violations.

6.1 Summary Cards

Card Description
Total Agents All agents in the tenant
Compliant Agents fully meeting current policy requirements
Grace Period Agents that were compliant when the policy changed and are being given time to adapt
Non-Compliant Agents actively violating the current policy
Suspended Agents that have been suspended due to persistent non-compliance

6.2 Agent Table

The compliance table shows all agents with: - Agent Name and operational status - Agent Type - LLM Provider and Model - Compliance Status badge (Compliant / Grace Period / Non-Compliant / Suspended) - Grace Period Countdown — A live timer showing hours remaining in the grace period

6.3 Status Filters

Tabs above the table filter by compliance status: All Agents, Compliant, Grace Period, Non-Compliant, Suspended.

6.4 Agent Compliance Detail

Clicking View on any agent opens a modal with:

  • Full agent identifiers (name, ID)
  • Status badge
  • Provider, model, and agent type
  • Grace period countdown (if applicable)
  • Violations list — Each violation shows the field and a human-readable description of what rule is being broken
  • Compliance History — A chronological timeline of past compliance status changes for this agent, including violation counts and grace period history

6.5 Policy Change History

The Change History button opens the same policy change history modal available in Agent Management, showing all tenant policy changes chronologically with before/after diffs.