API Keys

Manage provider API keys. Saving a key auto-discovers available models.

Gemini Active
Claude Active
Chatgpt Active
Grok Active
Deepseek Active
Openrouter Active

Feature → Model Assignments

Assign which AI model handles each feature. Use powerful models for complex tasks and fast/cheap models for simple ones.

🧠 Heavy Tasks (need reasoning & emotional intelligence)

Deep Coaching Conversations grok/grok-4.6
Extended coaching dialogues about goals, struggles, strategy
Escalation Messages grok/grok-4.6
Crafting escalating accountability messages
Goal Refinement (GROW) claude/claude-sonnet-5
Helping users define and refine goals using GROW model
Weekly Review Analysis claude/claude-opus-5
In-depth weekly progress analysis and planning

⚡ Light Tasks (summarization, classification, simple generation)

Behavioral Pattern Detection gemini/gemini-3.7-flash
Detecting recurring behaviors, excuses, and triggers
Check-In Responses gemini/gemini-flash-lite-latest
Quick AI responses after daily check-ins
Daily Reminder Text gemini/gemini-flash-lite-latest
Morning reminder message generation
Milestone Celebrations claude/claude-sonnet-5
Generating personalized milestone celebration messages
Mood & Sentiment Analysis gemini/gemini-flash-latest
Analyzing user mood trends and emotional patterns
Schedule Parsing grok/grok-4.6
Extracting structured times from free-text schedule descriptions
Session Summaries chatgpt/gpt-5.6-luna
Summarizing coaching sessions for memory
Smart Touchpoint Messages grok/grok-4.6
Contextual proactive messages throughout the day (nudges, checks, wraps)

🤖 Multi-Agent / Agentic (deep orchestrated analysis)

90-Day Plan (Multi-Agent) grok/grok-4.20-multi-agent-0309
Build a structured 90-day roadmap with milestones, weekly targets, daily tasks
Analyze My Day (Multi-Agent) grok/grok-4.20-multi-agent-0309
Comprehensive day analysis from audio transcripts, conversations, and goals
Document Drafting grok/grok-4.20-multi-agent-0309
Draft contracts, emails, proposals from conversation context
Goal Conflict Resolution grok/grok-4.20-multi-agent-0309
Resolve competing goals and schedule conflicts with multi-agent analysis
Plan My Week (Multi-Agent) grok/grok-4.20-multi-agent-0309
Deep weekly planning using multi-agent orchestration — calendar, goals, patterns

🎙️ Assistant, Voice, Calls & Skills

Everything else the coach runs on — realtime voice, phone-call missions, the capability planner that decides which skill to use, the calendar and email assistants, transcription and memory. Assign these exactly like the ones above.

Calendar Assistant gemini/gemini-flash-latest
Parses natural-language calendar add/move/rename/delete requests into structured actions
Call Missions (judge + report) grok/grok-4.6
Multi-place call-around missions: judge each call against the success criteria, compose the final report message
Capability Planner (skill routing) gemini/gemini-flash-lite-latest
Selects exactly one catalog skill (or "none") for every non-small-talk chat/voice turn so the server can bind + execute from the user message. Fast, cheap routing model — borrowed email_assistant before this dedicated key existed.
Council Dispatch Router gemini/gemini-3.6-flash
Classifies chat messages: delegate to KoraCouncil (and pick a mode) or keep in normal coaching.
Council Result Verifier claude/claude-opus-5
Audits a finished Council result against the user's original verbatim request before delivery.
Daily Email Brief grok/grok-4.6
Morning inbox brief (needs-you bullets + promo count)
Email Assistant grok/grok-4.6
Summarizes email search results for chat replies ("check my email for X")
Email Composer claude/claude-sonnet-5
Drafts emails for the draft-approve flow (chat/voice/call)
Feedback acknowledgment grok/grok-4.6
Short coach reply when the user rates a message (thumbs up/down)
Inbox Cleanup Triage gemini/gemini-flash-latest
Categorizes inbox mail into archive/read/keep for batch-approved cleanup
Insight Miner A (primary) gemini/gemini-3.6-flash
Coaching-insight extraction (model A) - runs every 5 min on each consolidated session
Insight Miner B (cross-check) grok/grok-4
Coaching-insight extraction (model B) - runs in parallel with A, agreement scored via embeddings
Live Voice Conversation Model grok/grok-voice-think-fast-2.0
xAI realtime speech-to-speech model for voice mode (/ws/voice). think-fast-2.0 = faster (0.7s to first audio) + smarter, $0.08/min. grok-voice-latest auto-tracks xAI upgrades
Memory Fact Extraction gemini/gemini-flash-latest
Extracts durable facts + intentions from every typed/voice/audio turn (Gemini-only: JSON mode). THE memory LLM.
Phase A Ice Breaker gemini/gemini-flash-latest
First-session intake: name, vulnerability admission, first anchor, permission framing
Phase B Life Chapters gemini/gemini-pro-latest
McAdams life chapter structuring + critical event capture
Phase C Dresser Drawers gemini/gemini-pro-latest
Per-drawer conversation using OARS micro-skills + permission gating
Phase D Deep Excavation gemini/gemini-pro-latest
Clean Language on metaphor, externalization, childhood linking
Proactive Scanner grok/grok-4.6
The every-few-minutes scan of genuinely-new user content that decides whether the coach reaches out (drop / act / call mission / to-do proposal). Analytical slot, strict JSON on the xAI endpoint.
Question Multiplier gemini/gemini-flash-latest
Generates 3-7 candidate follow-up questions after each user response, fans across dimensions
STT Per-Chunk Transcription grok/grok-stt
Engine for voice notes + live mic chunks. grok/grok-stt = default ($0.10/hr). chatgpt/gpt-transcribe = higher accuracy, $0.27/hr, no diarization. gemini/gemini-3.5-transcribe = verbatim, diarization + word timestamps, $0.30/hr. Non-Grok engines share the daily-hours ceiling and fall back to Grok
STT Speaker Diarization gemini/universal-3-5-pro
Whole-session diarization engine (consolidator 2nd pass) + per-chunk STT fallback model. assemblyai/universal-3-5-pro = default ($0.23/hr, 2h windows). gemini/gemini-3.5-transcribe = verbatim + diarization + word timestamps, $0.30/hr, 20-min windows
Self-Criticism Wrapper gemini/gemini-flash-latest
Reviews draft AI responses against anti-robotic / anti-solution rubric before delivery
Transcript Formatter (2nd pass) gemini/gemini-flash-latest
Whitespace-only re-layout of voice-note transcripts (verse lines, lists, paragraphs). Fidelity-guarded in code — words can never change
Wiki Page Compiler gemini/gemini-flash-latest
Self-maintaining wiki: compiles a user memory page from its cited sources (background batch — mid-tier is fine)

Bot Settings

Configure coaching behavior, reminders, and user preferences.

Deployment

Auto-deploy from GitHub pushes or trigger manual deploys.

GitHub Webhook Secret Set
Webhook URL: /coach/api/deploy
Pushes to main auto-deploy. Other branches ignored. 
Loading...
Saved!