Global Settings
Global settings are installation-wide runtime values that apply to every workspace; they are saved without restarting the API.
Where to find them
Most settings are on Management → System. Viewing requires View Settings; changing requires Manage Settings (administrators always have access). Host-related settings are on Hosts → Settings, and model and Bash policy settings are on Developer Tools → PolygentCode.
| Location | Tabs / sections |
|---|---|
| Management → System | General, Branding, Insights, Merge Conflicts, Code Review, Verification, Round Table, Plan, License, About |
| Hosts → Settings | Worker approval, worktree pool, preserved worktrees, IP restrictions |
| Developer Tools → PolygentCode | Settings (models and backends), User Settings (agent defaults), Permissions (Bash policy), Skills, Subagents |
Runtime settings are stored in {StoragePath}/settings.json; include it in backups. Secret values in it are encrypted with the installation's data-protection keys.
General
The General tab controls ticket execution and the global AI budget.
Tickets queue
| Setting | Default | Description |
|---|---|---|
| Pause / Resume (Ticket Execution) | Running | Pauses automatic ticket starts in every workspace. Queued tickets and approved plans keep their stage, priority, and queue order. Running work and ordinary conversations continue. Also available in the Tickets page header. |
| Low Priority Execution Hour (Local) | 2 | Hour of the day (0–23) when low-priority ticket work may run — starts, rejection resumes, hooks, reports, and follow-up messages. Low work runs after High and Normal work. |
Ticket concurrency is configured per session host under Hosts (see Per-host concurrency). A queued ticket starts only on an online, enabled host that allows its workspace, can run its model, and has a free ticket slot. To raise throughput, add hosts or raise their limits.
Global AI cost budget
| Setting | Default | Description |
|---|---|---|
| Global Monthly Budget (USD) | empty | Organization-wide monthly cap across all workspaces. Empty means no global cap. |
| Warning Threshold (%) | 80 | Percentage of any applicable limit (global, workspace, or ticket) at which a warning is sent. |
When a global budget is set, reaching it always blocks new work; the Block when global budget exceeded switch does not currently make the limit advisory. See AI Cost Budgets.
AI job settings
Each AI background job has its own tab with an enable switch (where applicable), a model, and a Tools editor that limits the MCP servers, Polygent MCP tools, custom subagents, skills, and agent tools the job may use. An unset model uses the Default (Daily use) tier.
| Tab | Setting | Default | Description |
|---|---|---|---|
| Insights | Enable Insights | On | Extract insights when a Develop session reaches Done. See Insights. |
| Insight Extractor | Default tier | Model used for extraction; a fast, low-cost model is sufficient. | |
| Merge Conflicts | Manual Merge Conflicts | Off | Skip AI resolution; every conflict goes straight to the review editor. Hides the model and tools. |
| AI Conflict Resolver | Default tier | Model used for AI conflict resolution. | |
| Tools | Built-in defaults | Capabilities available to the resolver. | |
| Code Review | Enable Code Review | Off | Review Develop-session changes when the agent finishes. See Code Review. |
| Code Reviewer | Default tier | Model used for review. | |
| Tools | Read, Glob, Grep | Capabilities available to the reviewer. | |
| Verification | Enable Verification | Off | Check Develop-session changes against the request when the agent finishes. See Verification. |
| Verification Agent | Default tier | Model used for verification. | |
| Tools | Read, Glob, Grep, Bash | Capabilities available to the verifier. | |
| Round Table | Tools | Built-in defaults | Capabilities available to round-table personas. |
| Plan | Tools | Built-in defaults | Capabilities available to ticket Plan generation. |
The code-review secret scan for committed credentials always runs alongside the AI review when Code Review is enabled and cannot be configured.
Default model tiers
Four system-wide tiers let users choose by task difficulty while administrators control the concrete model and reasoning effort.
| Tier | Intended use | When not configured |
|---|---|---|
| Max (Challenges) | Hard problems | Uses the Default (Daily use) model |
| High (Complex tasks) | Complex work | Uses the Default (Daily use) model |
| Default (Daily use) | Routine work | Uses the first available model |
| Instant | Fast work | Uses the Default (Daily use) model |
Configure tiers under Developer Tools → PolygentCode → Settings → Model Configuration. Models with configurable reasoning expose their effort levels, and the model and effort are saved together.
Every model selector lists the four tiers first, followed by direct model choices. A saved tier resolves to its current model and effort when work starts; direct model selections stay pinned. A tier inherits the availability and feature restrictions of its configured model — for example, a tier set to a model without conversation-fork support cannot be selected for the Planner. Unavailable tiers stay visible but disabled so the configuration problem is visible.
Branding, License, and About
- Branding — logo and accent color. See Branding.
- License — installed license status and upload. See Licensing.
- About — installed version and release notes; include it when contacting Support.
Hosts settings
The Hosts → Settings tab controls worker admission, worktree reuse, and network restrictions. Requires Manage Hosts.
| Setting | Default | Description |
|---|---|---|
| Require Admin Approval for Workers | On | New Session and Deployment Workers register disabled and receive no work until an administrator enables them on the Hosts page. |
| Enable Idle Worktree Reuse Pool | Off | Keep released working copies so new sessions on the same starting branch can reuse them (including build output and dependencies). |
| Idle Worktree Pool Size (per workspace) | 3 | Maximum idle working copies kept per workspace (0–100); the least recently used is evicted first. |
| Idle Worktree Max Age (days) | 7 | Idle working copies older than this are removed. |
| Preserved Worktree TTL (days) | 7 | How long a working copy preserved after an abnormal stop (uncommitted or unpushed work) is kept before automatic cleanup. 0 disables automatic cleanup. |
| IP Restrictions → Session Workers | empty | Allowed source IPs or wildcard patterns (for example 192.168.*.*) for Session Worker connections. Empty allows all. |
| IP Restrictions → Deploy Workers | empty | Same, for Deployment Worker connections. |
IP restrictions apply to all workers of that type; there is no per-key IP list.
Per-host concurrency
Each session host has its own limits, edited from the host's Concurrency action on the Hosts page.
| Limit | Default |
|---|---|
| Max Concurrent Tickets | 3 |
| Max Concurrent Sessions | Seeded from the worker's MaxConcurrentSessions on first registration |
| Max Concurrent Merge AI Processes | 2 |
| Max Concurrent Hook Tasks | 2 |
Each value must be a whole number of at least 1. The four values are saved together, so a failed save leaves the previous values unchanged. Changes apply to new work immediately; admitted work keeps its capacity. Other host actions: Allowed workspaces (restrict which workspaces the host serves; empty allows all), Set low priority (use the host only when normal hosts are full), and enable/disable.
Bash permissions
The administrator Bash denylist blocks shell commands in every Polygent Code session, in addition to the built-in safety floor that is always on.
Manage it under Developer Tools → PolygentCode → Permissions → Blocked Bash Commands. Viewing requires View Settings; saving requires Manage Settings. Each rule is a pattern plus a reason; a matching command is rejected before it runs and the agent receives the reason.
| Pattern form | Example | Matches |
|---|---|---|
| Substring | git push | Any command containing git push (case-insensitive). |
Glob (* / ?) | git push* | Unanchored wildcard match anywhere in the command. |
Regex (/.../) | /^rm\s+-rf/ | Case-insensitive regular expression. Invalid expressions are rejected on save; an expression that takes too long to evaluate blocks the command. |
Rules are global and apply to every workspace. Repository permissions.deny rules are enforced too; see Harness → Bash command policy.
Settings in appsettings.json
Some installation-wide values are not editable in the UI and are set in the API's appsettings.json or environment, then applied by restarting the API.
| Key | Default | Purpose |
|---|---|---|
ClientUrl | https://localhost:5173 | Public base URL used in links (pull requests, notifications, automations) and MCP OAuth callbacks. Must be changed for production. |
McpUrl | {ClientUrl}/mcp | MCP endpoint URL given to agents. See MCP Server. |
ShowFreeModels | false | List models with no usage charge in model dropdowns. |
ContentAssist:DefaultLanguage | Hebrew | Default language for Translate and Explain when a user's Native language is Use application default. One of Arabic, English, Hebrew, Russian; any other value prevents startup. |
The full list is in Configuration Overview.
See also
- Models & Backends — model catalog, tiers, backends, and custom models
- System Logs — logging configuration
- Authentication — OAuth and token settings
- Environment Variables — workspace-level values