Settings
Resources
AI provider connections
AI Providers
Claude
Connected
Anthropic · api.anthropic.com
API key configured · Last verified 2 min ago
OpenShift AI
Not configured
Red Hat · No cluster URL set
Token not configured
Ollama
Not running
Local · localhost:11434
Start Ollama to enable local model inference
Ramalama
Not running
Local · localhost:8080
Red Hat local model runner
Sandbox Runtime
OpenShell
Running
local-default · Podman · v2.1.0 · Uptime 2h 14m
3 active sandboxes · 3 gateways registered
Gateway
Each gateway is an independent runtime. Select one to view and manage its settings.
Gateways
local-default
Podman · local
team-shared
MicroVM · remote · OIDC
dev-sandbox
Podman · remote · mTLS
local-docker
Docker · local
local-default
local · Podman · v2.1.0 · Port 17670
Delete local-default? This will stop the process and remove the registration. This cannot be undone.
Running
· Uptime: 2h 14m
Driver
Set at creation — cannot be changed without recreating.
Endpoint
Sandboxes
3 running across all projects
Providers
Credential stores attached to this gateway
Inference routes
LLM provider endpoints available to sandboxes
Security logs
OCSF events from sandboxes on this gateway
Attach provider from vault
To create a new provider, go to Secret Vault →
Inference routes
LLM provider endpoints this gateway exposes to sandboxes. The gateway proxies requests and injects credentials automatically — agents call a stable local URL and never see raw API keys.
Add inference route
Driver Configuration — Podman
Some changes require recreating the gateway to take effect.
Supervisor
Supervisor image
The supervisor process runs inside every sandbox on this gateway and enforces security policies. It connects back to the gateway via an outbound gRPC stream — no inbound port is opened in the sandbox.
ghcr.io/nvidia/openshell/supervisor:0.0.71
Security enforcement
These policies are applied by the supervisor to all sandboxes on this gateway. Disabling any layer reduces isolation — only do so temporarily for debugging.
Filesystem isolation Landlock
Restricts sandbox filesystem access to declared paths only. Strict mode denies all unlisted paths; permissive mode logs violations instead.
Process isolation seccomp
Filters syscalls available to sandbox processes using the default seccomp profile. Reduces the kernel attack surface.
Network policy proxy OPA
Intercepts outbound HTTP/HTTPS traffic from sandbox processes. In enforce mode, requests that do not match an attached provider's allow-list are blocked and credentials are injected for matching hosts. In audit mode, violations are logged but not blocked.
Session & Defaults
SSH session TTL
Inactive SSH sessions are closed after this many seconds. Active sessions are unaffected. Applies to all sandboxes on this gateway.
s
Service forwarding
Allows sandboxes to expose services running on loopback to the host browser via the supervisor relay. Required for previewing web UIs, Jupyter notebooks, or any local service an agent starts inside the sandbox.
Log level
Verbosity of gateway and supervisor logs. Use debug only when diagnosing issues — it produces significant output.
Max concurrent sandboxes
Hard limit on simultaneously running sandboxes across all agents on this gateway. Set to 0 for unlimited.
Default CPU limit
CPU cores allocated per sandbox when the session request does not specify a limit. Applies across all drivers.
cores
Default memory limit
Memory in MiB allocated per sandbox when the session request does not specify a limit.
MiB
Auth & Token Issuance
The gateway issues short-lived JWTs for session auth, signed with an Ed25519 key pair managed by Kaiden.
Token TTL
Lifetime of issued JWT tokens. Shorter values reduce exposure; longer values reduce re-auth frequency.
s
Signing key
Ed25519 key pair used to sign gateway tokens. Rotating invalidates all active sessions.
kid:a3f8c2d1
Key path
On-disk path to the PEM private key. For remote gateways, configure directly on the server.
Observability
Metrics endpoint
Exposes Prometheus-compatible metrics at /metrics. Useful for Grafana, Datadog, or any compatible scraper.
Health check port
Port for /healthz and /readyz HTTP endpoints, used by container orchestrators and load balancers.
OPA Policies
Network allow-list rules enforced by the supervisor proxy inside each sandbox. Sandboxes can only reach hosts explicitly permitted here.
Allow-list
12 rules · 3 domain allow-entries · 2 prefix matchers
Gateway config (TOML)
Edit directly — changes are applied only when you click Apply Config.
Add gateway
Driver — fixed at creation
Podman (container-managed)
On Windows, the gateway runs as a Podman container managed by Kaiden. No driver selection needed.
Auth
Startup — local gateways only
Auto-start gateway
Start local gateways automatically when Kaiden launches. Remote gateways are always-on and unaffected.
Gateway port
Local port the gateway listens on (default 17670). Change only if there is a conflict.
Startup
Gateway container
On Windows, the gateway runs as a managed Podman container. Startup is handled automatically by Kaiden — no configuration needed.
Running
Gateway
The runtime that provisions sandboxes, enforces policy, and routes inference traffic.
Demo:
local-default
Managed
Managed · Podman · v2.1.0 · grpc://localhost:17670
Referenced gateway — configuration is managed externally and cannot be edited here.
Driver
Set at creation — cannot be changed without recreating the gateway.
Podman
Endpoint
gRPC address used by Kaiden to communicate with this gateway.
grpc://localhost:17670
Sandboxes
Active sandboxes running on this gateway.
3 running across all projects
Inference routes
LLM routing rules for sandboxes on this gateway.
Security logs
Sandbox egress events, OPA decisions, token activity.
Inference routes
Sandboxes call
http://inference.local/… — the supervisor proxies to the configured provider.New inference route
Driver — Podman
Some changes require recreating the gateway to take effect.
Supervisor
Image
ghcr.io/nvidia/openshell/supervisor:0.0.71
Landlock (filesystem isolation)
seccomp syscall filter
OPA network proxy
Session & Defaults
Service forwarding (sandbox→host)
Auth & Token Issuance
kid:a3f8c2d1
Observability
OPA Policies
Network allow-list enforced by the supervisor proxy inside each sandbox. 3 domain rules · 12 prefix matchers.
Gateway config (TOML)
Edit directly — changes apply only when you click Apply Config.
All sandboxes run on local-default.
Active gateway
local-default
Managed
Podman · v2.1.0 · grpc://localhost:17670
Switch to Single mode to interact with tabs — in the real UI, active gateway tabs are always interactive.
Other gateways
Sandboxes
Default behaviour for new sandboxes — overridable per project or per sandbox
Network
Allow known services
New sandboxes can reach services covered by their attached credentials. Additional hosts can be added per project.
Strict mode by default
Block all outbound traffic unless explicitly listed. Overrides "Allow known services" when on.
Filesystem
Include working directory
Automatically grant read-write access to the project working directory. Disable to create a fully isolated sandbox with no host filesystem access.
Isolation mode
Best effort falls back silently on older kernels. Strict fails sandbox creation if kernel-level isolation cannot be guaranteed.
Lifecycle
Keep sandbox after session ends
When on, the sandbox stays running after the agent session completes so you can reconnect or start a new session without recreating the environment.
Process identity
Unix user the agent process runs as inside the sandbox.
Logs
Security event stream from all sandboxes — OCSF v1.7.0 structured format
Recent security events
6m ago
TCP connect to auth.example.com:443 — no matching network policy
doc-updater
DENY
12m ago
POST api.anthropic.com/v1/messages
doc-updater
ALLOW
18m ago
GET api.github.com/repos/acme-corp/api
frontend-app
ALLOW
34m ago
Process started — claude-code agent launched
api-gateway
ALLOW
1h ago
Policy reloaded — revision #4 applied
doc-updater
ALLOW
Log settings
Minimum severity
Only show events at or above this level.
OCSF export sink
Stream structured events to an external endpoint (SIEM). Leave blank to disable.
Proxy
HTTP/HTTPS proxy configuration for outbound connections
HTTP Proxy
Proxy server for HTTP traffic (e.g. http://proxy.corp.com:3128)
HTTPS Proxy
Proxy server for HTTPS traffic. Leave blank to use HTTP Proxy value.
No Proxy
Comma-separated list of hostnames or CIDR ranges that bypass the proxy (e.g. localhost,127.0.0.1,.corp.com)
Use system proxy
Read proxy settings from the operating system environment (HTTP_PROXY, HTTPS_PROXY, NO_PROXY)
CLI Tools
Registered CLI binaries used by Kaiden. Override paths if binaries are installed in non-standard locations.
Registered Binaries
kdn
v0.9.2 installed
OpenShell
v2.1.0 installed
Claude
Configure the Anthropic Claude provider extension
Custom Base URL
Override the Claude API endpoint (
Leave blank to use the default:
claude.factory.baseURL).Leave blank to use the default:
https://api.anthropic.com
API Key
Your Anthropic API key (
Stored securely. Equivalent to setting
claude.factory.apiKey).Stored securely. Equivalent to setting
ANTHROPIC_API_KEY.
OpenAI-compatible
Configure a custom OpenAI-compatible provider endpoint
Base URL
The base URL of the OpenAI-compatible API endpoint (
Example:
openai.factory.baseURL).Example:
http://localhost:8000/v1
API Key
API key for the provider (
openai.factory.apiKey). May be left blank for unauthenticated local endpoints.
OpenShift AI
Configure the Red Hat OpenShift AI provider extension
Cluster API URL
The OpenShift cluster API endpoint (
Example:
openshiftai.factory.url).Example:
https://api.cluster.example.com:6443
Token
OpenShift bearer token for authentication (
Obtain via
openshiftai.factory.token).Obtain via
oc whoami -t or the OpenShift web console.
Preferences
Agent Workspace
Runtime
Override the container runtime used when creating agent workspaces.
Appearance
Appearance
Select between light or dark mode, or use your system setting.
Zoom Level
Set the zoom level. Positive values zoom in, negative values zoom out. Use decimals for fine control.
Navigation Bar Layout
Show icon and title, or just the icon in the navigation sidebar.
Chat
Show Chat Window
Show or hide the in-app chat window.
Disabled
Max Attachment File Size
Maximum file size in MB for chat attachments. Files exceeding this limit are rejected.
Editor
Font Size
Editor font size, in pixels.
Exit On Close
Quit on Close
Quit the app when the close button is clicked instead of minimizing to tray.
Disabled
Extension: kdn
Binary Path
Custom path to the kdn binary. Leave blank to use the default PATH location.
Extensions
Auto-update Extensions
Automatically download and apply extension updates when available.
Enabled
Check for Updates on Startup
Check for extension updates each time the application starts.
Enabled
Feedback dialog
Dialog
Show feedback dialog for experimental features.
Enabled
Kubernetes
kubeconfig Path
Path to the kubeconfig file. Defaults to ~/.kube/config.
Watch Contexts
Monitor Kubernetes context changes and update connection status in real time.
Enabled
Minimize on login
Minimize to Tray on Login
Start minimized to the system tray instead of opening the main window on login.
Disabled
Onboarding
Show Onboarding on Startup
Display the onboarding wizard when Kaiden starts for the first time.
Enabled
Start on login
Start on Login
Automatically launch Kaiden when you log in to your system.
Disabled
Tasks
Show Task Notifications
Show desktop notifications when background tasks complete or fail.
Enabled
Auto-dismiss Completed Tasks
Automatically remove completed tasks from the task list after a short delay.
Enabled
Telemetry
Enable Telemetry
Help improve Kaiden by sharing anonymous usage statistics. No personal data or code is sent.
Enabled
Telemetry Dialog Shown
Whether the telemetry consent dialog has already been displayed.
Disabled
Terminal
Font Size
Terminal font size, in pixels.
Line Height
Line height multiplier for the terminal. Minimum 1.0, maximum 4.0.
Scrollback Buffer
Number of lines to keep in the terminal scrollback buffer.
Update
Auto-check for Updates
Automatically check for new Kaiden releases on startup.
Enabled
Download Updates Automatically
Download updates in the background and prompt to install when ready.
Disabled
User Confirmation
Require Confirmation for Bulk Actions
Show a confirmation dialog before performing bulk operations (e.g. deleting multiple items).
Enabled
Window
Remember Window Position
Restore the window to its previous position and size on launch.
Enabled
Show in Taskbar
Display the Kaiden window in the taskbar / dock even when minimized to tray.
Enabled
Gemini
Configure the Google Gemini provider extension — gemini.factory.*
Enter your Gemini API key. Get one at aistudio.google.com
Mistral
Configure the Mistral AI provider extension — mistral.factory.*
Enter your Mistral API key. Get one at console.mistral.ai
Vertex AI
Connect to Anthropic models on Google Cloud Vertex AI — vertex-ai.factory.*
Project ID where Anthropic models are enabled (not the project name or number).
Google Cloud region where the model endpoint is deployed (e.g. us-east5, europe-west1).
Path to gcloud credentials file. Leave blank to use gcloud auth application-default login.
Milvus
Configure the Milvus vector store connection used for knowledge base embeddings — milvus.*
Display name for this Milvus connection.
Docling
Configure the Docling document chunking provider — docling.*
Display name for this Docling chunk provider connection.
Experimental Beta
Enable or disable experimental features. These may be unstable or change without notice.
Enable experimental features
Activate early-access capabilities across all Kaiden extensions and providers. Individual features may have their own toggle below once enabled.
Agent Feed
Show the Agent Feed panel in the sidebar with live agent activity, alerts, and progress across all running workspaces.
Debug logging
Write verbose debug logs to the developer console. Useful for diagnosing connection issues.