Kucatoo-Code · Wiki

Kucatoo-Code Wiki

Kucatoo-Code is an open-source AI IDE console and model harness, built China-first: it targets the Chinese model ecosystem (Kimi, DeepSeek, GLM, Qwen, MiniMax) through their international API endpoints, wrapping them in a Flask web app backed by an async agent harness with tools, memory, workflows, and skills. The web app serves a 4-column workspace (models / prompt / response / capability console) with direct-mode and agent-mode queries, voice (TTS, dictation, STS voice chat), and per-workspace cost tracking.

For LLM agents: this index is the entry point. Read the one topic page relevant to the question rather than scanning code; each page declares the code paths it describes in its sources: front matter and is updated diff-driven (see "Maintaining this wiki" below). Deeper design rationale lives in ../../guidance-library/Model-Harness_*.md — the wiki links, the library explains.

Pages

Page Scope
Architecture Processes/ports (:5000 app, :5001 supervisor, :5002 voice WS), directory map, query lifecycles, auth
Provider mesh 10 registered models, endpoints/keys (Bailian plan vs direct), failover, circuit breaker, cost accounting
Agents ReAct vs Reflective loops, modes & policies, write guard, compaction, budgets, ask_user
Tools ToolSystem architecture, 6 builtins + 46 ported tools, confirmation/read-only policy (auto-generated table)
Memory & workflows STM/MTM/LTM memory, scheduler (cron/interval/webhook/event), SkillDAG sidecar
Voice & audio TTS providers, STS voice chat, wake-word personas, streaming speech, dictation
Web UI & API Pages, console tabs, SSE event reference, usage stats/TSV export, guests (auto-generated route table)
Configuration Env vars, ports, data directories, .env loading
Hosting Client micro-VM tiers (Micro–Pro, slim/full, yyz USD pricing), tier toggles, central pay hub, provisioning path, upgrade/termination lifecycle

Conventions

Maintaining this wiki

verified against code: 2026-08-09