Local configuration, environment variables, directories, capabilities, model keys, and tool settings for nenjo run.

Worker Config

nenjo run loads config from .nenjo/config.toml, then applies environment variable and CLI overrides.

By default .nenjo is ~/.nenjo. You can point the worker somewhere else with --nenjo-dir or NENJO_DIR.

Local Directories

Path	Purpose
`.nenjo/config.toml`	Serialized worker config
`.nenjo/workspace`	Project worktrees, synced repositories, and task workspaces
`.nenjo/state`	Memory, sessions, session content, and crypto state
`.nenjo/manifests`	Cached bootstrap manifest resources
`.nenjo/gitconfig`	Worker-specific git config when `GITHUB_TOKEN` is set
`.nenjo/git-credential-helper.sh`	Worker-specific GitHub token helper when `GITHUB_TOKEN` is set

The state directory is the important backup target for local worker-generated state.

Required Identity

The worker requires a Nenjo API key:

export NENJO_API_KEY="..."
nenjo run

The key can also be stored as api_key in .nenjo/config.toml.

CLI and Environment Overrides

Setting	CLI flag	Environment variable	Default
Backend API URL	`--backend-url`	`NENJO_API_URL`	`https://api.nenjo.ai`
NATS URL	`--nats-url`	`NATS_URL`	Bootstrap value, then `tls://nats.nenjo.ai`
Log level	`--log-level`	`RUST_LOG`	`info`
Show log target	`--log-target`	`NENJO_LOG_TARGET`	`false`
Config directory	`--nenjo-dir`	`NENJO_DIR`	`~/.nenjo`
Worker capabilities	`--capabilities`	`NENJO_CAPABILITIES`	All capabilities
Harness display name	`--harness-name`	`NENJO_HARNESS_NAME`	unset
Harness labels	`--harness-labels`	`NENJO_HARNESS_LABELS`	none

Most workers should not set NATS_URL. The bootstrap response supplies the event bus settings when available.

Model Provider Keys

Model keys can be stored in [model_provider_api_keys] or supplied by environment variables. Environment variables win.

Provider	Environment variable
OpenAI	`OPENAI_API_KEY`
Anthropic	`ANTHROPIC_API_KEY`
OpenRouter	`OPENROUTER_API_KEY`
Ollama	`OLLAMA_API_KEY`
Groq	`GROQ_API_KEY`
Venice	`VENICE_API_KEY`
Mistral	`MISTRAL_API_KEY`
DeepSeek	`DEEPSEEK_API_KEY`
xAI	`XAI_API_KEY`
Together	`TOGETHER_API_KEY`
Fireworks	`FIREWORKS_API_KEY`
Perplexity	`PERPLEXITY_API_KEY`
Cohere	`COHERE_API_KEY`
Moonshot	`MOONSHOT_API_KEY`
GLM	`GLM_API_KEY`
Minimax	`MINIMAX_API_KEY`
Google/Gemini	`GOOGLE_AI_API_KEY` or `GEMINI_API_KEY`
OpenAI-compatible	`OPENAI_COMPATIBLE_API_KEY`

For OpenAI-compatible providers, the manifest can use a tagged provider name such as openai-compatible:sambanova. The worker first looks for a config key named sambanova, then SAMBANOVA_API_KEY, then the generic OpenAI-compatible key.

Capabilities

Capabilities limit which command families this worker handles.

nenjo run --capabilities chat,task,repo

Valid values are chat, task, cron, manifest, repo, and ping. If no capabilities are configured, the worker runs in full mode and handles all of them.

Tool Config

These sections control optional built-in tools:

Config section	Default	Purpose
`[web_search]`	enabled with DuckDuckGo	Adds web search; can use Brave with `brave_api_key`
`[web_fetch]`	disabled	Adds web page fetching with allowed and blocked domains
`[http_request]`	disabled	Adds general HTTP requests with allowed domains
`[browser]`	disabled	Adds browser open and screenshot tools
`[memory]`	`markdown`	Selects the memory backend
`[git]`	unset	Sets git author, committer, and optional SSH signing key for worker git operations

Web fetch and HTTP request both have response-size and timeout settings. Domain filters are applied before requests are made.

Autonomy and Reliability

[autonomy] controls the application-level safety policy for local actions. Defaults keep work inside the workspace, block high-risk commands, require approval for medium-risk shell commands, and include a blocklist for destructive or system-level commands.

[reliability] controls model retry behavior:

Setting	Default
`max_retries`	`2`
`backoff_ms`	`500`
`fallback_providers`	none
`model_fallbacks`	none

The worker's provider registry wraps created model clients with the reliability layer.

Security and Audit

[security.sandbox] can auto-detect or explicitly choose a sandbox backend: auto, landlock, firejail, bubblewrap, docker, or none.

[security.audit] defaults to enabled and writes audit.log under the .nenjo directory. Audit events can also be HMAC-signed when configured.

Example

api_key = "..."
harness_name = "local-dev"
harness_labels = ["macbook", "repo-work"]
capabilities = ["chat", "task", "repo", "manifest"]

[model_provider_api_keys]
anthropic = "..."
openai = "..."

[web_search]
enabled = true
provider = "duckduckgo"
max_results = 5
timeout_secs = 15

[web]
allow_private_hosts = false

[web_fetch]
enabled = true
allowed_hosts = ["docs.rs", "github.com"]
blocked_hosts = []

[git]
user_name = "Nenjo Worker"
user_email = "worker@example.com"

Worker Config

On this page