Config Schema

Configuration uses the same validated keys in user ~/.logician/settings.json and trusted project .logician.json files. Unknown keys produce diagnostics rather than silently changing runtime behavior.

The current flat runtime keys include:

{
  "baseUrl": "http://127.0.0.1:8080",
  "model": "model-id",
  "memoriam": {
    "mode": "sdk",
    "python": "/path/to/memoriam/.venv/bin/python",
    "args": ["-m", "memoriam.integration.sdk_worker"],
    "failOpen": true,
    "timeoutMs": 30000,
    "config": { "db_path": "~/.logician/memories.db" }
  },
  "reasoner": "none",
  "reasonerConfig": {},
  "thinkingLevel": "off",
  "inferenceMode": "none"
}

reasoner defaults to none. A selected reasoner runs before the ordinary tool-capable agent loop and its reasonerConfig values override that reasoner's registry defaults.

thinkingLevel defaults to off. inferenceMode defaults to none, labeled Provider in the UI, which omits Logician's sampling presets and lets the provider use its defaults.

memoriam wires the standalone memoriam memory engine in as an out-of-process JSON-lines SDK worker (same pattern as legroom). Omit it or set "mode": "off" to run without persistent memory. failOpen (default true) keeps the agent turn alive if the worker is unreachable; config.db_path is the SQLite database path.

Common settings

{
  "baseUrl": "string",
  "model": "string",
  "maxTokens": "positive number",
  "temperature": "number from 0 to 2",
  "permissionMode": "plan|ask|acceptEdits|acceptAll",
  "permissions": { "allow": ["string"], "deny": ["string"] },
  "inferenceMode": "auto|none|thinking-general|thinking-coding|instruct-general|instruct-reasoning|instruct-coding|deterministic|creative|analytical",
  "thinkingLevel": "off|minimal|low|medium|high|xhigh",
  "executionProfile": "autonomous|minimal",
  "toolExecution": "parallel|sequential",
  "compaction": {
    "enabled": "boolean",
    "reserveTokens": "positive number",
    "keepRecentTokens": "positive number"
  },
	"legroom": {
		"mode": "off|sdk",
		"python": "python3",
		"args": ["-m", "legroom.sdk_worker"],
		"failOpen": "boolean (default true)",
		"timeoutMs": "positive number (default 30000)",
		"config": {}
	},
  "maxParallelAgents": "positive number",
  "mcpServers": {
    "name": { "command": "string", "args": ["string"], "enabled": true }
  },
  "allowedPaths": ["absolute path"],
  "allowAllPaths": "boolean",
  "lsp": { "enabled": "boolean", "timeoutMs": "positive number" },
  "truncation": {},
  "plugins": {}
}

Environment variable overrides

Environment overrides exist for selected deployment-sensitive settings:

Config path	Environment variable
baseUrl	LOGICIAN_LLM_URL
model	LOGICIAN_MODEL
systemPrompt	LOGICIAN_SYSTEM_PROMPT
contextWindowTokens	LOGICIAN_CONTEXT_WINDOW
hooks	LOGICIAN_HOOKS

Logician reads ~/.logician/.env before resolving MCP servers. It is intended for secrets referenced as ${VARIABLE} in MCP headers or env maps.