Configuration

Kandev has three distinct configuration surfaces:

• backend startup configuration: defaults, config.yaml, then environment variables;
• persistent product settings edited in the web UI and stored in the database; and
• executor, agent, repository, and workflow profiles stored through their own Settings pages.

This page is the startup-configuration reference. Executor-specific fields are covered in Executors, and deployment examples are in Docker, Kubernetes, and Run as a service.

Load order and lifecycle

At backend startup, later sources override earlier ones:

1. embedded production-profile values and built-in defaults;
2. the first readable config.yaml; and
3. environment variables.

The public launcher has no --config option. It searches for a file named exactly config.yaml in the backend working directory and then /etc/kandev/. A missing file is allowed. Malformed YAML, unmarshal errors, and validated invalid values stop startup.

Configuration is read at process start; there is no file watcher. Restart Kandev after changing YAML or environment variables. The CLI, desktop shell, service manager, Docker, or Kubernetes may set environment variables on the backend, so those values can override a file unexpectedly. Use the process/service/container environment as the final source of truth.

Environment-variable naming

Viper maps a nested YAML key by replacing . with _, adding KANDEV_, and uppercasing. It does not split camelCase words. For example:

database.dbName       -> KANDEV_DATABASE_DBNAME
logging.maxSizeMb     -> KANDEV_LOGGING_MAXSIZEMB
repoClone.basePath    -> KANDEV_REPOCLONE_BASEPATH

Some common camelCase keys have explicit compatibility aliases. Use the documented exact names below; snake_case spellings not listed here are not equivalent.

Complete backend reference

Root and server

The default host exposes the server on every interface even though the CLI prints a localhost URL. The current local product path must not be treated as an authenticated multi-user perimeter. For remote access, bind to loopback and use a trusted authenticated tunnel/proxy, or isolate the network at the deployment layer.

Database

SQLite is the supported default and enables WAL mode. PostgreSQL deployments must provision the database, network policy, TLS trust, backups, and credentials before starting Kandev. Passing the password in an environment variable avoids putting it in YAML but still exposes it to processes/administrators allowed to inspect the environment; use your platform's secret injection controls.

database.path is an advanced override. Persistence honors it, but the current Settings → System → Database and Backups services derive their displayed path, WAL files, backup source, and restore destination from <home>/data/kandev.db. Those UI operations are therefore not reliable for a custom SQLite location. Use operator-managed backup/restore for a custom path and verify the actual database path printed at startup.

Database-only snapshots also omit <home>/data/master.key, the AES-256 key used to decrypt stored secrets. Preserve that owner-only key with an independently secured home/data backup; restoring the database without its matching key leaves encrypted credentials unreadable. See Operations.

Event bus and NATS

An external NATS URL moves event traffic across the configured network and can embed credentials/TLS parameters. Protect it as a secret where applicable, require TLS for untrusted networks, and keep namespaces distinct when deployments share one NATS server. clusterId does not provide isolation in the current implementation.

Docker runtime

The Docker socket is effectively root-equivalent on many hosts. Do not publish it or assume docker.tlsVerify secures a TCP daemon—it currently does not. Configure TLS through a supported Docker endpoint/environment and validate it independently, or keep the daemon local. See Docker and Executors.

Core agent service

The launcher starts agentctl, performs a one-time nonce handshake, and supplies the resulting per-launch token internally. Do not persist or proxy its bootstrap/auth state. Agent command, model, environment, permission, and MCP configuration belongs in agent profiles rather than this section.

Authentication, Office, Plugins, voice, and feature flags

Do not infer security from auth.jwtSecret: setting it currently does not turn the local server into an authenticated public service. Office's JWT key has a narrower, active purpose. Store both active secrets and third-party API keys in your deployment secret manager; never commit them in config.yaml.

The voice fallback sends audio to the configured OpenAI transcription service and incurs that provider's network, data-handling, and billing behavior. Browser-native speech recognition has its own browser/vendor behavior and does not use this server key.

Logging

The format default becomes JSON when KUBERNETES_SERVICE_HOST is non-empty or KANDEV_ENV is exactly production/prod; otherwise it is text. Rotation settings apply only to file output. Active log files are created owner-only (0600) on Unix, so a log shipper running as another user cannot read them without an explicit permission design.

Debug output may contain repository paths, subprocess output, prompts, file content, and tool-call data. Treat it as sensitive.

Repository, worktree, and clone paths

Discovery roots bound automatic filesystem traversal, so scope them narrowly. They do not authorize explicitly selected repository paths: Add Local Repository validates and saves the exact accessible Git repository the user chooses without widening automatic scans. Worktrees and clones can contain credentials or generated files ignored by Git; review repository copy-file and setup/cleanup settings before remote execution. See Git operations.

Debug configuration

Debug mode is high risk. It enables local diagnostic surfaces and implies KANDEV_DEBUG_AGENT_MESSAGES=true and KANDEV_DEBUG_PPROF_ENABLED=true when not explicitly locked by the environment. ACP JSONL frames include complete prompts, file content, and tool calls. Do not enable it on a shared or network-exposed backend.

Minimal examples

For a local-only CLI server with an isolated home:

KANDEV_SERVER_HOST=127.0.0.1 \
KANDEV_HOME_DIR="$PWD/.kandev-local" \
kandev --headless

For a file-based deployment, override only what is needed:

homeDir: "/srv/kandev"

server:
  host: "127.0.0.1"
  port: 38429

logging:
  format: "json"
  outputPath: "/var/log/kandev/backend.log"

repositoryDiscovery:
  roots:
    - "/srv/repositories"

This example changes automatic discovery only. An explicitly selected repository may live outside /srv/repositories if the Kandev process can access and validate it.

A complete shape, including compatibility fields, is:

homeDir: ""

server:
  host: "0.0.0.0"
  port: 38429
  readTimeout: 30
  writeTimeout: 30
  webInternalUrl: ""

database:
  driver: "sqlite"
  path: ""
  host: "localhost"
  port: 5432
  user: "kandev"
  password: ""
  dbName: "kandev"
  sslMode: "disable"
  maxConns: 25
  minConns: 5

nats:
  url: ""
  clusterId: "kandev-cluster" # compatibility-only today
  clientId: "kandev-client"
  maxReconnects: 10

events:
  namespace: ""

docker:
  enabled: true
  host: "unix:///var/run/docker.sock" # use the Windows named pipe on Windows
  apiVersion: ""
  tlsVerify: false                    # compatibility-only today
  defaultNetwork: "kandev-network"  # compatibility-only today
  volumeBasePath: "/var/lib/kandev/volumes" # compatibility-only today

agent:
  standaloneHost: "localhost"
  standalonePort: 39429

auth:
  jwtSecret: ""       # compatibility-only for the main HTTP product path
  tokenDuration: 3600 # compatibility-only for the main HTTP product path

logging:
  level: "info"
  format: "text"
  outputPath: "stdout"
  maxSizeMb: 100
  maxBackups: 5
  maxAgeDays: 30
  compress: true

repositoryDiscovery:
  roots: []                # automatic scan roots; explicit paths need not be included
  maxDepth: 5

worktree:
  enabled: true
  defaultBranch: "main"    # compatibility-only today
  cleanupOnRemove: true    # compatibility-only today
  fetchTimeoutSeconds: 60
  pullTimeoutSeconds: 60

repoClone:
  basePath: ""

debug:
  devMode: false
  pprofEnabled: false

office:
  jwtSigningKey: ""

voice:
  openAIApiKey: ""

features:
  office: false
  plugins: false

Copying this entire file is unnecessary and can freeze old defaults in a deployment. Keep only deliberate overrides. On Windows, do not copy the Unix Docker host/path literals from this example.

Runtime feature toggles

Settings → System → Feature Toggles manages three startup-time flags:

UI changes are persisted in the database and require a restart. An explicitly set environment value wins and locks the UI control. Otherwise a database override wins over the embedded profile/default. Resetting a toggle removes its database override.

The source checkout's make dev activates the embedded development profile, which enables Office, debug surfaces, ACP logging, and a mock agent. Installed run/desktop builds select the safe production profile unless the environment explicitly opts in. E2E mock variables and routes are test-only and must never be enabled on a public deployment.

Credentials and product settings

Most integrations, executor profiles, agent profiles, MCP servers, repository settings, and UI preferences are persistent database records edited under Settings. They are not fields in config.yaml. Secret values use an encrypted secret store backed by <home>/data/master.key; filesystem permissions, database backups, and key backup are part of the security boundary.

For headless injection, Kandev can also read agent credentials from the process environment by their required name (for example ANTHROPIC_API_KEY) or the KANDEV_-prefixed form. KANDEV_CREDENTIALS_FILE adds a fallback JSON provider:

{
  "ANTHROPIC_API_KEY": "replace-at-deployment-time",
  "OPENAI_API_KEY": "replace-at-deployment-time"
}

The file is loaded lazily, expects a flat string-to-string object, and is cached; restart after changing it. A missing file behaves as no file credentials, while unreadable or invalid JSON produces credential-resolution errors. The database secret store and environment providers are consulted before this file. Restrict file permissions to the Kandev service account and never commit it.

Profile environment variables are eventually injected into agent subprocesses or remote executor environments. Anyone who can edit a profile, inspect a remote host/container, enable debug frame logs, or run commands as the Kandev account may be able to access them. Use least-privilege, task-scoped credentials and rotate them after exposure.

Advanced operator tuning

These startup-only variables are supported by specific runtime components but are not YAML fields. Leave them at defaults unless diagnosing a measured problem.

Changing concurrency/queue values trades memory and process pressure against throughput. Values are captured during process/component initialization and need a restart.

The current OTLP exporter strips an http:// or https:// prefix from the configured endpoint and always uses WithInsecure(). Treat this as implementation-bound cleartext transport: send it only to a trusted private collector over a protected network, not directly across an untrusted network. The service name is kandev-agentctl, and spans can include task/session/execution IDs plus raw agent-event JSON truncated to 8192 characters. That payload can contain prompts, files, and tool data. Use collector-side access controls and retention accordingly.

ACP debug-log controls

These apply only when KANDEV_DEBUG_AGENT_MESSAGES=true:

Retention values must be positive integers; invalid/non-positive values use defaults. Directories and files use owner-only 0700/0600 modes on Unix. Rotation, age pruning, and file-count pruning bound normal growth, but these files remain highly sensitive and can exist inside a Docker executor rather than on the host.

Validation and troubleshooting

Startup validation currently enforces:

• server.port: 1-65535;
• database.driver: sqlite or postgres;
• PostgreSQL port, non-empty user/database name, and supported SSL mode;
• positive auth.tokenDuration;
• logging level/format; and
• positive repositoryDiscovery.maxDepth.

Other fields can pass configuration validation and still fail later—for example an unreachable NATS/PostgreSQL/Docker endpoint, unwritable log path, nonsensical timeout, or incompatible pool sizes. A field appearing in the schema does not prove its subsystem is available.

If a value appears ignored:

1. confirm the exact environment spelling, especially camelCase keys;
2. inspect the launcher/service/container environment for an overriding value;
3. confirm config.yaml is in the backend working directory or /etc/kandev/;
4. restart the backend; and
5. check whether the field is marked compatibility-only above.

Use kandev --verbose to surface startup errors. Do not use --debug merely to diagnose a YAML typo on an exposed machine; verbose logs are usually sufficient.

Variables used only to assemble/test the runtime—such as KANDEV_WEB_DIST_DIR, KANDEV_DESKTOP_RUNTIME_DIR, mock/E2E switches, supervisor socket/manifest values, and bootstrap nonces—are internal implementation contracts, not supported deployment configuration. KANDEV_BUNDLE_DIR is the narrow exception documented for installer/package integration in CLI; end users should still let the installer set it.