AI Runtime

Configure the model providers and command surfaces that power THOR. API keys live in server env only and are never shown.

Runtime status

Provider router · approval-gated command mode
kimi
Default Provider
openai
Fallback Provider
Active
Model Router
0
Model Runs
0
Fallback events
0
Failed runs
—
Avg latency
—
Last run
Kimi / Moonshot
Connected
Base URL: https://api.moonshot.ai/v1
Default model: kimi-k2.6
Use case: Long-context reasoning, strategy, analysis
API key: configured (hidden)
OpenAI / ChatGPT
Not Connected
Base URL: https://api.openai.com/v1
Default model: gpt-4o-mini
Use case: General reasoning, ChatGPT bridge, fallback
API key: set OPENAI_API_KEY in server env
NVIDIA NIM
Not Connected
Base URL: https://integrate.api.nvidia.com/v1
Default model: meta/llama-3.3-70b-instruct
Use case: Cost-efficient hosted open models
API key: set NVIDIA_API_KEY in server env

ChatGPT Connector (MCP)

Approval-gated command surface
MCP Endpoint
/mcp
Command mode
Approval-Gated
Allowed tools (9)
  • • thor_create_squad_mission
  • • thor_get_squad_status
  • • thor_list_blockers
  • • thor_get_approval_requests
  • • thor_request_qa
  • • thor_run_gmail_response_squad
  • • thor_run_google_ads_audit_squad
  • • thor_create_codex_task
  • • thor_get_operational_picture
Blocked actions
  • ✕ Direct Gmail send
  • ✕ Direct Google Ads changes
  • ✕ Direct database writes
  • ✕ Secret / token access
ChatGPT requests route through /api/thor/command as source=chatgpt. High-risk commands wait for approval; nothing bypasses the Google gateway, QA or external-action approvals.

Model run logs

Provider · model · task · status · latency · tokens
No model runs yet. Use “Test” above to generate one.