[ LOCAL-FIRST · DESKTOP OPENCLAW PLATFORM ]
XiaoyaoClaw is a local-first desktop application platform: create AI bots in one window, connect 9 major IM channels like Feishu, WeChat and Telegram, bring your own keys (BYOK) and switch models freely — configuration, sessions and data all stay on your machine.
FIG.01 — XiaoyaoClaw desktop console: multi-channel access / always-on agents / fully local data
SEC. 01
PLATFORM / 6 MODULES
ALL LOCAL-FIRST
LOCAL-FIRST / NO CLOUD LOCK-IN
All configurations, sessions and states are stored in the local data directory — no cloud lock-in. Works offline; data sovereignty is fully yours — critical when handling internal docs and customer conversations.
SLACK / DISCORD / TG / Feishu / WeChat / DingTalk / QQ / WeCom / WA
OAuth-guided onboarding with real-time connection monitoring. Wherever your customers chat, your AI assistant is there — one bot, all channels online.
MULTI-INSTANCE PER CHANNEL
Connect multiple Feishu apps and multiple Slack workspaces at once. Isolated by department or customer group, independently controlled, no interference — single-instance configs still supported.
OPENAI / ANTHROPIC / GOOGLE / …
Bring your own API keys to connect directly to model providers, or use the official hosted option. Bind model pools per bot and switch freely by cost and performance — budget under control.
SKILL MARKETPLACE
Install community skill packs with one command: office collaboration, dev & coding, creative media, research & analysis. The 9 official ecosystem skills below work out of the box.
SESSIONS & HISTORY
Filter conversation records by bot, channel and status. Markdown rendering and session reset supported. Every interaction is traceable and auditable.
SEC. 02
9 open-source skills form a complete loop around the XiaoyaoClaw agent workspace: settle in → get work done → remember → audit → measure. All MIT licensed, zero dependencies, dual-platform (Windows / macOS), installable via ClawHub with one command.
9 SKILLS / MIT
ZERO-DEP / DUAL-OS
workspace-initializer
Give every agent a "home": 7 standard directories + WORKSPACE.md spec. No memory loss on restart, clean directories, no conflicting configs.
workspace-auditor
Read-only workspace health check: directory compliance, task health, memory gaps, KB orphans and junk files — 5 categories in one run.
task-progress-tracker
Directories as containers, PROGRESS.md as progress: sessions hand off cleanly, documents stay findable, history is traceable — pure file system, zero overhead.
agent-orchestrator
Dispatch in one sentence → parallel fan-out → transparent progress → summarized with sources → automatic retry on failure. Multiple always-on agents collaborate like a team.
kb-retriever
Layered index navigation + progressive retrieval over local files (md / pdf / xlsx). No vector DB, no cloud calls — answers cite sources for verification.
usage-report
Reads local session JSONL directly: how long each task took, which tool was slowest, how many tokens each model ate. Optimize agents with data, not vibes.
web-clipper
Send a URL, get a clean local Markdown: full frontmatter, CJK-friendly, batch dedupe — directly indexable by your knowledge base.
memory-distill
Distill long conversations into structured memory: core entries go to MEMORY.md, daily entries to per-day logs. Incremental dedupe prevents bloat; sensitive info auto-skipped.
commander
Let any Agent Skills tool — Claude Code, Codex, OpenCode, Trae, DSH and more — directly command your XiaoyaoClaw multi-agent system.
$ One-line install of ecosystem skills (CLAWHUB)
$ clawhub install xiaoyaoclaw-workspace-auditor $ clawhub install xiaoyaoclaw-task-progress-tracker $ clawhub install xiaoyaoclaw-agent-orchestrator $ clawhub install xiaoyaoclaw-kb-retriever # …same for the rest; works on Windows / macOS
SEC. 03
The 9 skills aren't isolated tools — they're a pipeline that wraps around the agent workspace:
Auditor runs periodic read-only health checks on the entire workspace — messy directories, stalled tasks, broken memory chains — and delivers a graded report with fix suggestions.
Usage Report lays out duration, tool bottlenecks and token bills in plain numbers — quantify the payoff of every improvement.
Cross-tool Commander opens the whole system up to external agents like Claude Code / Codex — wherever you work, command from there.
SEC. 04 — SPEC SHEET
SEC. 05 — DEPLOY
Download the XiaoyaoClaw desktop client — three steps to set up: connect a channel → bind a model → install skills. All ecosystem skills are open-source (MIT) and freely composable.
Multi-agent deployment / workspace standardization / custom skill development / memory system setup
