Server Settings
Configure global server settings and integrations
Failed to load settings
Portal Access and Branding
Give business users a dedicated workspace for their assigned reports and apps.
Also turn on Portal for each environment under Environments → Portal and give users access. Environment settings
Portal
Application connections
Portal shows connected applications, such as IBM i programs, through a separate CoderFlow address. The applications keep running on their own systems.
Set up each application under Environments → Portal → Applications. These addresses apply to all environments.
General Settings
Ask CoderFlow
Answer Feedback
Loading feedback totals…
Usage Telemetry
Captures completed inquiries (question, answer, metadata) and pushes them to a processing server to measure and improve answer quality. Off by default; transcripts are only captured while on. Answer ratings are always saved locally and are sent when telemetry is enabled.
AI Providers
Task execution defaults
Set the server-wide model defaults for tasks. Environment, template, and personal overrides may select a different task model.
Automatic model selection (Auto tiers)
Auto rates each task low, medium, or high and runs it on the model set here for that tier. Choose the same model for two tiers to limit the range. Models picked explicitly for a task or environment are not affected.
Codex speed modes
Utility AI
A single server-wide model for short AI operations performed outside task agents.
Agent Keepalive Settings
Configure automatic token refresh for each agent to keep them authenticated and responsive
Claude
Codex
Gemini
Bob
Grok
Git Providers
Configure Git providers for repository authentication. Environments can reference these providers to access private repositories.
Setup Repository
Email Settings (SMTP)
Test Configuration
Slack Integration
Connect a Slack bot so users can @mention CoderFlow, run the generated slash command to create tasks, use message shortcuts to create tasks or save threads as objectives, and use its link command to connect their Slack identity.
Quick Setup: Create Slack App from Manifest
- Go to api.slack.com/apps and click Create New App → From an app manifest
- Select your workspace, click Next, choose the JSON tab
- Copy the manifest below and paste it in
- Click Next, review, and Create
- Copy Signing Secret from Basic Information, and Bot Token from OAuth & Permissions (install the app first)
The manifest uses the Slack Webhook Base URL when set, otherwise the Task Link Base URL or current server URL.
Channel Mappings
Map Slack channels to specific environments. Messages in mapped channels auto-select the configured environment.
Test Configuration
Microsoft Teams Integration
Connect a Microsoft Teams agent so users can mention CoderFlow or message it directly to create tasks. Inbound requests are authenticated with the Microsoft 365 Agents SDK.
Quick Setup: Teams App Manifest
- Create an Azure Bot / Microsoft app registration and set its public HTTPS messaging endpoint to
- Paste the Microsoft App ID and client secret below
- Click Download App Package to get a ready-to-upload
.zip(manifest + placeholder icons), or copy the manifest to build your own package - Upload the app in Teams for your tenant or team
The manifest uses Task Link Base URL when set, otherwise the current server origin. Bump App version (higher x.y.z) so Teams treats a re-upload as an update rather than requiring you to delete and re-add the app. The downloaded package includes placeholder color.png (192×192) and outline.png (32×32) icons — replace them with your own branding before publishing.
id) that identifies the app in the store — a distinct GUID from the Microsoft App ID above. Keep it stable across versions so updates apply in place instead of creating a duplicate app. If you already published the app, set this to its existing Teams App ID.
Channel Mappings
Map Teams channels or conversations to specific environments. Easiest: in Teams open the channel → … → Get link to channel, then paste that URL into a mapping's link box below to auto-fill the Team ID and Channel ID. You can also enter the IDs manually.
Test Configuration
Jira Integration
Connect to Jira so users can import issues as tasks and post results back as comments.
Test Configuration
MCP Server
Expose CoderFlow as a Model Context Protocol (MCP) server so external agents such as Claude Desktop / Cowork or ChatGPT can hand work to CoderFlow directly: create tasks, pass along a session handoff, check status, send follow-ups, and launch judge / review runs on a task group. Each connection authenticates with a user's personal API key and runs under that user's permissions.
/mcp and the OAuth token endpoints, for example the public address of the dedicated MCP listener. Leave blank to use the Site URL. Login and consent always use the Site URL.
MCP_INGRESS_PORT (and optionally MCP_INGRESS_HOST) to serve only /mcp, the token and registration endpoints, and the OAuth discovery documents on a separate port.
The dedicated listener is enabled but no Site URL is set in General settings. OAuth clients would be sent to the listener for login, where the authorization page does not exist. Set the Site URL to the main application address.
*.example.com matches subdomains. Client registrations and authorizations that redirect elsewhere are refused. Leave empty to allow any host (the MCP default). Loopback addresses used by local clients are always allowed.
GET/DELETE /api/mcp-oauth/grants.
Connect a client
Hosted clients (claude.ai, Cowork): with OAuth enabled, add the endpoint URL as a custom connector; the user signs in to CoderFlow in the browser and approves access. No key to copy.
Local clients can use OAuth too, or a personal API key sent as a Bearer token. Keys are issued by coder login, coder login --sso, the VS Code extension sign-in, or POST /api-keys. Replace <API_KEY> below.
Authentication
Behind a reverse proxy serving HTTPS (Cloudflare, nginx, load balancer)? Enable
Trust Proxy in General settings
so OAuth callback URLs are sent as https://.
Automatically create user accounts when someone logs in via SSO for the first time.
Access binding assigned to auto-provisioned users. Can be changed later per user.
Show username/password form alongside SSO button. Disable for SSO-only authentication.
Require a time-based one-time password (TOTP) from an authenticator app for CoderFlow local logins. Applies to browser sessions and CLI/VS Code API keys. OIDC logins remain governed by your identity provider.
When enabled, enrolled users are challenged for a code at login.
Optional: users may enroll; enrolled users are challenged at login. Required: users must enroll before they can complete local login.
Shown in the user's authenticator app next to their account.
Enabling Required mode needs a fresh code from your own authenticator (you must be enrolled first). This prevents locking everyone out.
TOTP parameters: 6 digits · 30-second period · SHA-1 · ±1 step clock skew
CA Certificates
Add custom CA certificates for containers to trust internal/corporate certificate authorities.
Container Resource Limits
Contain each task, interactive session, deployment, test, documentation publish, and prompt container. CPU and memory limits are ceilings; CPU shares are proportional scheduling weights under contention. They do not reserve resources or guarantee fixed CPU time.
Applies CPU weight plus CPU, memory, and process ceilings to every new runtime container. Existing containers keep the settings they started with.
Hard ceiling per container. A hard cap can reduce a single task's burst performance on an otherwise idle host, because it is not work-conserving. Leave blank to let deploy, test, and docs jobs burst, and to give task containers half the Docker host's cores — never below two, where in-container build parallelism collapses to serial. Configuring a shared cgroup parent bounds the whole fleet instead and removes the per-container ceiling.
Relative CPU scheduling weight when runnable containers contend. Docker's usual default is 1,024; the conservative 256 default makes task workloads yield to higher-weight peers. Deploy, test, and docs jobs take half this weight, since they burst freely and must yield just as freely. Shares are proportional, not a fixed CPU allocation.
Per-container ceiling. Leave blank to preserve the existing default of half the host memory. Measure peak per-agent memory before lowering it; a ceiling that is too low can terminate a task.
Maximum number of processes the container may run, from 32 to 65,536.
Max Concurrent Agents is an active-agent admission limit, not a bound on all running containers, and it does not affect CPU sizing. Interactive, deployment, test, prompt, documentation, and lingering post-turn containers can increase the running-container count.
Automatic Cleanup
Automatically reclaim Docker disk space when the filesystem holding Docker's storage comes under pressure. Only dangling images, unused build cache, and unused networks are removed. Tagged images, containers, and volumes are never removed automatically — use the cleanup actions on the Server Administration page for those.
Removes dangling images, unused build cache, and unused networks when disk pressure is detected. Idle task containers are already stopped and reclaimed separately, once their changes have been pushed.
How full Docker's filesystem must be before cleanup runs. Both levels remove the same things; only the threshold differs.
Skip cleanup unless at least this much space can be reclaimed, so a full disk with nothing to delete does not prune repeatedly.
Update Management
Configure how the server can be updated and restarted from the Administration page.
When enabled, admins can apply updates and restart the server from the Health tab on the Server Administration page.
Off by default. Allows Server Admins to upload and install server .tgz packages when Enable Web Updates is also on. Install only trusted deployment outputs: package inspection and checksums do not authenticate the publisher, and installed code runs with the server account's privileges. Custom Update Commands are unsupported for local packages.
Command to install a new version. Use {version} as a placeholder for the target version number.
Command to restart the server process after an update. Leave empty to use SIGTERM and rely on an external process manager to bring the service back up.
Extra npm release channels (dist-tags) the Health tab can check and install from, separated by commas. The official Stable channel is always available. Leave empty to hide the channel selector and only offer official releases. Prerelease channels such as dev carry internal test builds.