Configuration Management

Managing models, providers, logging, and settings using the config engine

Tiny Claw uses a config engine backed by SQLite to manage all configuration. The tinyclaw config command provides a CLI interface for managing models, providers, and logging settings.

Configuration Architecture

Config Engine

SQLite-backed persistent storage for all settings. Located at ~/.tinyclaw/data/config.db.

Secrets Engine

Separate encrypted storage for API keys and tokens. Uses AES-256-GCM encryption bound to your machine.

The config engine stores references to secrets (key names), while actual secret values are encrypted in the secrets engine.

Model Configuration

Two-Tier Model Hierarchy

Tiny Claw uses a two-tier provider system:

Built-in Provider

Ollama Cloud — Free fallback, always available

• Default provider configured during setup
• Used when primary provider is unavailable
• Model can be switched between built-in options

Primary Provider

Plugin Provider — Optional, overrides built-in as default

• Set via provider plugins (OpenAI, Anthropic, etc.)
• Used as the default provider in the smart router
• Falls back to built-in if unavailable

View Current Model Configuration

Display the active model configuration for both tiers:

tinyclaw config model

Example Output

Model Configuration

Built-in (Ollama Cloud - always available as fallback)
  Model    : qwen2.5:32b-instruct
  Base URL : https://ollama.com/api

Primary (overrides built-in as default provider)
  Model    : gpt-4
  Base URL : https://api.openai.com/v1

The smart router uses Primary as the default provider.
Built-in is the fallback if Primary becomes unavailable.

List Available Built-in Models

See all models available for the built-in Ollama Cloud provider:

tinyclaw config model list

Example Output

Available Built-in Models

● qwen2.5:32b-instruct
  Best all-around model - balanced speed and capability

○ qwen2.5:14b-instruct
  Faster, smaller model for simple tasks

○ qwen2.5:72b-instruct
  Largest model for complex reasoning

Switch with: tinyclaw config model builtin <tag>

The ● marker indicates the currently active model.

Switch Built-in Model

Change the built-in provider’s model:

tinyclaw config model builtin <tag>

Example

$ tinyclaw config model builtin qwen2.5:14b-instruct

✔ Built-in model switched to qwen2.5:14b-instruct

Restart Tiny Claw for changes to take effect.

Available tags:

• qwen2.5:32b-instruct (default)
• qwen2.5:14b-instruct
• qwen2.5:72b-instruct

Changes require restarting Tiny Claw with tinyclaw start to take effect.

View Primary Provider

Check the current primary provider configuration:

tinyclaw config model primary

With Primary Set

Primary Provider

  Model      : gpt-4
  Base URL   : https://api.openai.com/v1
  API Key    : stored as "provider:openai"

Clear with: tinyclaw config model primary clear

No Primary Set

Primary Provider (not configured)

  No primary provider set. The built-in Ollama Cloud provider
  is used as the default for the smart router.

  To add a primary provider, install a provider plugin and
  ask Tiny Claw to set it as primary. You can also tell Tiny Claw:
  "list my providers" or "set OpenAI as my primary provider"

Clear Primary Provider

Remove the primary provider override and revert to built-in:

tinyclaw config model primary clear

Example

$ tinyclaw config model primary clear

✔ Primary provider cleared
  Built-in will be used as the default provider.

Restart Tiny Claw for changes to take effect.

Logging Configuration

View Current Log Level

Display the active log level:

tinyclaw config logging

Example Output

Log Level

○ debug
● info
○ warn
○ error
○ silent

Change with: tinyclaw config logging <level>
Override per session with: tinyclaw start --verbose

Set Log Level

Change the persistent log level:

tinyclaw config logging <level>

Available levels:

• debug — Show all logs including debug messages
• info — Normal operational logs (default)
• warn — Warnings and errors only
• error — Errors only
• silent — No logs

Example

$ tinyclaw config logging debug

✔ Log level set to debug

Restart Tiny Claw for changes to take effect.

You can override the log level for a single session using tinyclaw start --verbose (sets level to debug).

Setting Providers Conversationally

While the config command provides CLI access, you can also manage providers by talking to your Tiny Claw agent:

1. Ask to list providers

   "list my providers"
   

   Your agent will show all registered providers (built-in + plugins) and their status.

2. Set primary provider

   "set OpenAI as my primary provider"
   

   The agent will update the config to use OpenAI as the default provider in the smart router.

3. Switch models

   "switch to GPT-4o"
   

   The agent can reconfigure provider models on the fly.

This conversational approach is more natural than CLI commands — your agent handles the config engine operations for you.

Data Directory

All configuration is stored in the Tiny Claw data directory:

~/.tinyclaw/
├── data/
│   ├── config.db          # Config engine database
│   ├── config.db-wal      # SQLite write-ahead log
│   └── config.db-shm      # SQLite shared memory
├── heartware/             # Soul seed and identity
├── learning/              # Learned patterns
└── audit/                 # Audit logs

You can override the data directory with the TINYCLAW_DATA_DIR environment variable:

export TINYCLAW_DATA_DIR="/path/to/custom/dir"
tinyclaw config model

Configuration Persistence

All configuration changes are:

• Persisted immediately to the SQLite database
• Validated before being saved
• Require restart to take effect (the agent loads config at boot)

To apply changes:

1. Make changes

   tinyclaw config model builtin qwen2.5:72b-instruct
   tinyclaw config logging debug
   
2. Restart agent

   tinyclaw start
   

   The agent will load the new configuration on startup.

Next Steps

Learn how to manage LLM providers and switch models