Configuration
CortexPrism stores its configuration in ~/.cortex/config.json (created by cortex setup).
{
"version": 1,
"defaultProvider": "anthropic",
"providers": {
"anthropic": { "kind": "anthropic", "model": "claude-sonnet-4-5", "apiKey": "sk-ant-...", "reasoningEffort": "medium" },
"openai": { "kind": "openai", "model": "gpt-4o", "apiKey": "sk-...", "baseUrl": "https://api.openai.com/v1" },
"google": { "kind": "google", "model": "gemini-2.0-flash", "apiKey": "..." },
"ollama": { "kind": "ollama", "model": "llama3.2", "baseUrl": "http://localhost:11434", "numCtx": 8192, "keepAlive": "5m" }
},
"agent": {
"name": "Cortex",
"maxTurns": 50,
"streamOutput": true
},
"router": {
"enabled": false,
"strategy": "cascade",
"confidenceThreshold": 0.7
},
"modelSelection": {
"enabled": false,
"mode": "balanced",
"observeThreshold": 50,
"enforceConfidence": 0.85,
"suggestConfidence": 0.65,
"quartermasterProvider": "google",
"quartermasterModel": "gemini-2.0-flash",
"autoModelPool": [
{ "provider": "anthropic", "model": "claude-3-5-sonnet-20241022", "enabled": true },
{ "provider": "openai", "model": "gpt-4o", "enabled": true }
]
},
"supervisor": {
"provider": "google",
"model": "gemini-2.0-flash",
"cacheTTL": 3600
},
"computerUse": {
"enabled": false,
"display": { "width": 1920, "height": 1080 },
"runtime": "native",
"screenshot": { "format": "png", "quality": 90 },
"timeout": { "action": 5000, "display": 10000 },
"approval": { "requireApproval": true }
},
"sandbox": {
"timeoutMs": 30000,
"maxOutputBytes": 65536,
"scrollAmount": 3,
"dockerImages": {}
},
"voice": {
"enabled": false,
"provider": "openai",
"defaultVoice": "alloy",
"speed": 1.0,
"autoTTS": false
},
"update": {
"channel": "stable",
"checkOnStartup": true,
"autoUpdate": false,
"checkIntervalHours": 24,
"githubToken": null,
"gpgKeyPath": null
},
"pluginUpdate": {
"checkOnStartup": true,
"autoUpdate": false,
"checkIntervalHours": 24
},
"logging": {
"level": "error",
"fileEnabled": true,
"fileMaxBytes": 10485760,
"fileMaxFiles": 5
},
"webAuth": {
"requireAuth": true
}
}{ "kind": "anthropic", "model": "claude-sonnet-4-5", "apiKey": "sk-ant-...", "reasoningEffort": "medium" }reasoningEffort: low (1024 tokens) | medium (4096) | high (16384) — maps to thinking.budget_tokens
{ "kind": "openai", "model": "gpt-4o", "apiKey": "sk-...", "baseUrl": "https://api.openai.com/v1", "reasoningEffort": "medium" }o-series models use max_completion_tokens instead of max_tokens and omit temperature/top_p.
{ "kind": "google", "model": "gemini-2.0-flash", "apiKey": "...", "reasoningEffort": "high" }thinkingConfig.thinkingBudget mapped from reasoningEffort. Supports native vision via inlineData.
{ "kind": "ollama", "model": "llama3.2", "baseUrl": "http://localhost:11434", "numCtx": 8192, "keepAlive": "5m", "numThread": 4 }{ "kind": "bedrock", "model": "us.amazon.nova-pro-v1:0", "region": "us-east-1", "secretKey": "..." }Uses Converse API with inferenceConfig for maxTokens, temperature, and topP.
{ "kind": "openrouter", "model": "openai/gpt-4o", "apiKey": "...", "httpReferer": "https://myapp.com", "xTitle": "My App" }Injects HTTP-Referer and X-Title headers for OpenRouter rankings.
{ "kind": "perplexity", "model": "sonar-pro", "apiKey": "...", "searchRecencyFilter": "week", "returnCitations": true }searchRecencyFilter: month | week | day | hour. returnCitations and returnImages forwarded.
{ "kind": "together", "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo", "apiKey": "...", "repetitionPenalty": 1.1 }repetitionPenalty → repetition_penalty body field.
{ "kind": "litellm", "model": "gpt-4o", "apiKey": "...", "dropParams": true }dropParams → drop_params to silently ignore unsupported parameters.
{ "kind": "venice", "model": "default", "apiKey": "...", "includeVeniceSystemPrompt": true }includeVeniceSystemPrompt → venice_parameters.include_venice_system_prompt.
{ "kind": "deepinfra", "model": "meta-llama/Llama-3.3-70B-Instruct", "apiKey": "..." }Serverless GPU inference. Supports repetitionPenalty.
{ "kind": "hyperbolic", "model": "deepseek-ai/DeepSeek-V3", "apiKey": "..." }80% cheaper than traditional cloud providers.
{ "kind": "minimax", "model": "minimax-m3", "apiKey": "..." }MiniMax M3: 80.5% SWE-bench Verified at $0.30/$1.20 per 1M tokens.
{ "kind": "zhipu", "model": "glm-4-flash", "apiKey": "..." }Free tier available (glm-4-flash).
{ "kind": "replicate", "model": "meta/meta-llama-3-70b-instruct", "apiKey": "..." }Uses predictions-based REST API with polling (non-streaming) and SSE (streaming).
{ "kind": "cloudflare", "model": "@cf/meta/llama-3-8b-instruct", "apiKey": "...", "accountId": "..." }Edge inference via Cloudflare's AI platform. Requires both API token and Account ID. accountId → account_id path parameter.
{ "kind": "lmstudio", "model": "local-model", "baseUrl": "http://localhost:1234/v1", "numCtx": 4096, "keepAlive": "10m" }Uses OpenAI-compatible path with num_ctx and keep_alive forwarding.
| Variable | Purpose |
|---|---|
CORTEX_DATA_DIR |
Override data directory (default: ~/.cortex/data/) |
CORTEX_CONFIG_DIR |
Override config directory (default: ~/.cortex/) |
CORTEX_VAULT_KEY |
Vault decryption passphrase |
CORTEX_LOG_LEVEL |
Override log level (trace, debug, info, warn, error, silent) |
GITHUB_TOKEN |
GitHub personal access token |
GH_TOKEN |
Alternative GitHub token |
OPENAI_API_KEY |
OpenAI API key (alternative to config) |
CORTEX_API_URL |
Server URL for CLI login/API commands (default: http://localhost:11434) |
Multi-user authentication is enabled when the users table has at least one user. On first run,
an auto-admin is created from the vault password. API tokens provide persistent CLI authentication.
{
"webAuth": {
"requireAuth": true
}
}| Field | Default | Description |
|---|---|---|
webAuth.requireAuth |
true |
Require login to access the web UI |
User, team, token, and federation management is handled through the REST API and CLI (cortex users,
cortex teams, cortex login). See CLI Reference and Security.
Tries cheapest provider first, escalates on low confidence.
{
"router": {
"enabled": true,
"strategy": "cascade",
"confidenceThreshold": 0.7,
"cascade": [
{ "provider": "ollama", "model": "llama3.2:3b" },
{ "provider": "anthropic", "model": "claude-sonnet-4-5" }
]
}
}Routes based on prompt complexity scoring.
{
"router": {
"enabled": true,
"strategy": "threshold",
"confidenceThreshold": 0.5,
"threshold": {
"strongProvider": "anthropic",
"strongModel": "claude-sonnet-4-5",
"weakProvider": "ollama",
"weakModel": "llama3.2:3b",
"scorer": "heuristic"
}
}
}{
"modelSelection": {
"enabled": false,
"mode": "balanced",
"observeThreshold": 50,
"enforceConfidence": 0.85,
"suggestConfidence": 0.65,
"costBudget": 1.0,
"qualityThreshold": 0.7,
"allowedProviders": ["anthropic", "openai", "ollama"],
"quartermasterProvider": "google",
"quartermasterModel": "gemini-2.0-flash",
"autoModelPool": [
{ "provider": "anthropic", "model": "claude-sonnet-4-5", "enabled": true },
{ "provider": "openai", "model": "gpt-4o", "enabled": true }
]
}
}| Field | Default | Description |
|---|---|---|
mode |
"balanced" |
Strategy: conservative, balanced, or aggressive
|
observeThreshold |
50 |
LLM calls before MQM activates predictions |
enforceConfidence |
0.85 |
Threshold for enforce mode (override model) |
suggestConfidence |
0.65 |
Threshold for suggest mode (hint in prompt) |
autoModelPool |
[] |
Available models for Auto mode, each {provider, model, enabled} (v0.46+) |
quartermasterProvider |
— | Dedicated provider for MQM analysis |
quartermasterModel |
— | Dedicated model for MQM analysis |
{
"supervisor": {
"provider": "google",
"model": "gemini-2.0-flash",
"cacheTTL": 3600
}
}
}{
"logging": {
"level": "error",
"fileMaxBytes": 10485760,
"fileMaxFiles": 5,
"otlp": {
"endpoint": "http://localhost:4318"
},
"langfuse": {
"publicKey": "pk-lf-...",
"secretKey": "sk-lf-..."
}
}
}{
"computerUse": {
"enabled": false,
"display": { "width": 1920, "height": 1080 },
"runtime": "native",
"screenshot": { "format": "png", "quality": 90 },
"timeout": { "action": 5000, "display": 10000 },
"approval": { "requireApproval": true, "autoApproveReadOnly": false },
"docker": { "image": "cortexprism/computer-use:latest" }
}
}{
"sandbox": {
"timeoutMs": 30000,
"maxOutputBytes": 65536,
"scrollAmount": 3,
"dockerImages": {}
}
}
`runtime`: `docker` | `gvisor` (kernel-level syscall filtering via `runsc`) | `subprocess`.
## Server Config
- [LLM Providers](LLM-Providers) — All 30 supported providers
- [Model Routing](Model-Routing) — Router strategies
- [Model Quartermaster](Model-Quartermaster) — MQM learning system
- [Security](Security) — Security model
- [Observability](Observability) — Logging, metrics, and tracing