Skip to main content

Global Settings

Global settings control platform-wide scheduling, model jobs, worker admission, safety controls, branding, and budgets.

The Settings area exposes platform-wide configuration that is not tied to a single workspace. These values affect every workspace, every session, and every queue. Permissions: settings.view to read, settings.manage to change (admins always have access).

Where to Find It

Sidebar → Settings → Global. All values are persisted in the database (or in appsettings.json where noted) and take effect immediately, without restarting the API.

Tickets Queue

SettingDefaultDescription
Low Priority Execution Hour2Hour of the day (0–23, server local time) when low-priority ticket work is eligible to run, including starts, rejection resumes, hooks, reports, and follow-up messages. Low work runs after High and Normal work.

AI ticket concurrency is configured per host: each session host has its own Max Concurrent Tickets value, set on the Hosts page. For each queued ticket the gate looks for a host that can actually run it — workspace-allowlist-compatible, able to run the selected model, online, with at least one free slot under its own cap. If no host fits, the ticket waits. To raise overall throughput, enable more hosts or raise individual host caps; to dedicate a host to specific workspaces, set its allowlist.

Polygent Code Models

Backend connection API keys configured for Polygent Code are write-only in the UI, masked after saving, and protected at rest in the settings file.

Default Model

SettingDefaultDescription
Default ModelFirst available modelSystem-wide default model and, when supported, reasoning effort.

Pick a Default Model to have it automatically selected everywhere a model is chosen — Create Session, Bots, Plans, Round Tables, Automations, Start Ticket templates, and more. Models with configurable reasoning expose their available effort levels; the selected model and effort are saved together. Models without configurable effort remain directly selectable.

Several selectors also offer an explicit "Default" choice in the model dropdown (for example the Bot form and Start Ticket templates). Picking "Default" means follow the system-wide model and reasoning effort rather than pinning a specific combination.

If a saved model or effort is no longer available because the model was hidden, removed, or filtered by free-model visibility, settings require an available default and normalize stale values to the first available model. Job-specific model settings below override the global default when set; when left as Default, they inherit the system-wide model and effort.

Insights

SettingDefaultDescription
Insights EnabledtrueMaster switch for auto-extraction of insights from completed Develop sessions.
Insights ModeldefaultModel used to extract Insights. Pick a low-cost, fast model — extraction is short and high-volume.

Merge Conflicts

SettingDefaultDescription
Merge Conflicts ModeldefaultModel used for AI-assisted merge conflict resolution. Pick one strong at code reasoning.
Manual Merge ConflictsfalseWhen true, AI resolution is skipped and every conflict goes straight to the review diff editor.
ToolsDefaultsRestrict MCP servers, Polygent MCP tools, custom subagents, skills, and agent tools for AI conflict resolution. Hidden when manual mode is on.

Code Review

SettingDefaultDescription
Code Review EnabledfalseRun an automated code review pass on completed sessions.
Code Review ModeldefaultModel used for code review. Defaults to the workspace's session model when unset.
ToolsRead, Glob, GrepRestrict MCP servers, Polygent MCP tools, custom subagents, skills, and agent tools for separate code review runs.

When Code Review is enabled, two always-on deterministic scans run alongside the AI pass and are not configurable: a secret scan for committed credentials and a duplicate-code scan for functions copied from elsewhere in the codebase. Their findings are always included regardless of the model chosen. See Code Review for details.

Verification

SettingDefaultDescription
Verification EnabledfalseRun an automated verification pass when the agent finishes (before Code Review).
Verification ModeldefaultModel used for verification.
ToolsRead, Glob, GrepRestrict MCP servers, Polygent MCP tools, custom subagents, skills, and agent tools for separate verification runs.

Round Table

SettingDefaultDescription
ToolsBuilt-in defaultsRestrict MCP servers, Polygent MCP tools, custom subagents, skills, and agent tools for round tables.

Plan

SettingDefaultDescription
ToolsProvider defaultsRestrict MCP servers, Polygent MCP tools, custom subagents, skills, and provider tools for plans.

Hosts Settings

The following settings live on the Hosts page, not on the global Settings page. They control host-related behavior, capacity, agent approval, worktree pooling, and network allow-lists.

Per-Host Concurrency

Open a session host's Manage → Concurrency editor to set its maximum concurrent tickets, sessions, merge-AI processes, and hook tasks. Each value must be a whole number of at least 1. Saving applies all four values together, so a failed request leaves the previous configuration unchanged. Changes apply to subsequent work immediately without restarting the API; work already admitted continues under its previous capacity.

Agent Approval

SettingDefaultDescription
Require Admin Approval for AgentstrueWhen enabled, newly registered Deploy and Session Workers must be approved by an admin on the Hosts page before they can pick up work.

Worktree Pool

SettingDefaultDescription
Enable Idle Worktree Reuse PoolfalseKeep idle worktrees around so new sessions can grab a pre-cloned tree instead of cloning fresh.
Idle Worktree Pool Size3How many idle worktrees to keep per workspace when the pool is enabled.
Idle Worktree Max AgenoneIf set, idle worktrees older than this duration are evicted from the pool.

IP Restrictions

SettingDefaultDescription
Session WorkersemptyAllowed IPs / wildcard patterns (e.g. 192.168.*.*) that can connect to the session-worker hub. Empty = allow all.
Deployment WorkersemptySame, but for the deployment-worker hub.

Bash Permissions

The global blocked-bash-command rules live on the Polygent Code → Permissions tab (Sidebar → Dev Tools → Polygent Code) and let administrators block dangerous shell commands across every Polygent Code session. Each rule is a pattern plus an optional reason; when a session's shell-tool call matches a rule, the command is rejected before it runs and the agent receives the reason so it can adapt.

Pattern formExampleMatches
Substringgit pushAny command containing git push (case-insensitive).
Glob (* / ?)git push*git push, cd x && git push --force, etc.
Regex (/.../)/^rm\s+-rf/A command starting with rm -rf.

Rules defined here are global — they apply to all workspaces and every Polygent Code session. Requires the Manage Settings permission.

Read-Only (from appsettings.json)

These show in the Settings UI for visibility but are configured in appsettings.json and require an API restart to change.

SettingSourceDescription
Client URLappsettings.json: ClientUrl (or env ClientUrl)Base URL of the React client. Used for shared links, public bot URLs, and OAuth redirects. Falls back to the request URL if unset.
MCP URLappsettings.json: McpUrl (or env McpUrl)Public URL advertised for the MCP endpoint when generating agent MCP configurations.
Database Providerappsettings.json: Database:ProviderActive database provider (read-only here — see Database).

AI Cost Budgets

The global monthly spend cap and the warning threshold are configured under Settings → General → Global AI Cost Budget. When a global budget is set it is always enforced. These are documented in full in the AI Cost Budgets guide.

Branding

The custom logo and main accent color are configured under Settings → Branding. See the Branding guide.

Model Configuration

The model catalog appears in the Settings UI with model visibility toggles and custom models. See Models & Backends.

See Also