Configuration

CortexFlow-AI reads ~/.cortexflow/config.toml. Every section has sane defaults — most deployments only need [agent] and one [channels.*] block.

Minimal config

[agent]

[channels.telegram]
bot_token = "ENV:TELEGRAM_BOT_TOKEN"

Any string value prefixed with ENV: resolves to that environment variable at load time, so secrets never need to live in the file itself:

bot_token = "ENV:TELEGRAM_BOT_TOKEN"

Key	Default	Description
`name`	`"My Assistant"`	Display name used in status output and the system prompt
`model`	`"auto"`	`"auto"` enables task-aware routing; set explicitly to pin one model

Key	Default	Description
`primary`	`claude-opus-4-8`	Used for complex reasoning and code review tasks
`fallback`	`gemini-2.0-flash`	Used if the primary model fails or for general tasks
`fast`	`gemini-2.0-flash`	Used for cheap/quick inference tasks
`local`	`ollama/llama3.2`	Used in privacy mode — zero external API calls
`anthropic_api_key`	`""`	Supports `ENV:` resolution
`gemini_api_key`	`""`	Supports `ENV:` resolution
`deepseek_api_key`	`""`	Supports `ENV:` resolution
`ollama_base_url`	`http://localhost:11434`	Local Ollama server address

Key	Default	Description
`short_term_ttl`	`3600`	Redis TTL in seconds for active conversation context
`long_term_days`	`90`	Retention window for SQLite long-term memory
`redis_url`	`redis://localhost:6379`	Short-term memory backend
`qdrant_url`	`http://localhost:6333`	Semantic search backend
`sqlite_path`	`~/.cortexflow/memory.db`	Long-term memory database file

Key	Default	Description
`stt`	`whisper`	`whisper` or `none`
`tts`	`kokoro`	`elevenlabs`, `kokoro`, `system`, or `none`
`tts_voice`	`Rachel`	ElevenLabs voice name (or cloned voice ID)
`stt_model`	`base`	faster-whisper model size: tiny/base/small/medium/large-v3
`stt_device`	`cpu`	`cpu` or `cuda`
`tts_engine`	`kokoro`	Local fallback engine
`elevenlabs_api_key`	`""`	Supports `ENV:` resolution
`elevenlabs_voice_id`	`""`	Set automatically after `cortex voice clone`

Key	Default	Description
`port`	`7432`	Gateway listen port (WebSocket + REST)
`bind`	`127.0.0.1`	Use `0.0.0.0` to accept connections from other hosts/containers

Key	Default	Description
`web_port`	`3000`	Next.js web UI dev/serve port

Every channel block supports enabled (boolean) plus channel-specific keys — see Channels for the full per-platform reference.

[channels.discord]
enabled = true
bot_token = "ENV:DISCORD_BOT_TOKEN"