LyncoreYour desktop AI butler
An agent platform installed on your computer. Not a webpage, not an API — a digital twin you launch with a double-click.
Skip to content
Nine core features
This is what your AI companion should look like — Lyncore
💻
On your machine, not someone else's server
Lyncore is a desktop agent system, not SaaS, not a webpage. All your chats, documents, and agent configs live on your hard drive. Physically, nothing has to pass through any cloud server. Works offline (with local Ollama models).
Data sovereigntyFully offlineNo cloud dependency
🔒
Multi-layer safety, guardrails on duty
Python sandbox isolates dangerous modules (subprocess/socket/os all blocked); Shell has a 13-rule blacklist + regex filtering + CWD lock; file access whitelist; hallucination detection. Enterprise adds gVisor container-level isolation + audit logs + RBAC — safety everywhere!
Sandbox isolationCommand blacklistHallucination check
🧬
Gets smarter with use, not "dead" after setup
Some platforms' workflows are hand-configured; some platforms' bots must be built from scratch — both need manual setup. Lyncore's Evolution Engine watches your repeated actions and auto-drafts skill proposals. Lyncore's agents may be dumbest on day one, but by day one hundred they become the shadow that knows you best.
Auto-discoveryShadow previewHuman review
🖥️
Helps you control your computer, not just "chat"
Browser auto-search and scraping; desktop AI screenshots to understand the UI and click/type automatically. DAG workflows with visual orchestration: search 100 pages → scrape → compare → email, all in one line. RPA tools lack AI understanding; AI platforms lack desktop control — Lyncore has both.
Browser automationDesktop controlDAG orchestration
🔑
Bring your own model key, zero markup
Some agents take 30%+ between you and the model vendor. Big-tech agent platforms lock you into their ecosystem. Lyncore: paste in your own DeepSeek/Qwen key and the tokens you pay are between you and the model vendor — we never touch them.
Zero middleman cutUser-defined models
📁
Knowledge base bound to agents — your expert assistants
OpenClaw needs an external RAG. Dify has a KB but in the cloud. Lyncore: drop in your product manual/contracts/tech docs, vector index + BM25 hybrid retrieval, agents bind to it and carry it automatically — answers with quotes and sources.
Vector + full-textAgent bindingScoped access
🏢
Run teams and AI with the same rigor — no runaway bills
Lyncore Enterprise ships PostgreSQL RLS row-level isolation — Department A's agents can never see Department B's docs. Three-tier quota inheritance: tenant → department → user; each level's Token and sandbox-memory cap auto-cascades from the parent. CostGuard middleware hard-blocks at quota — when balance runs out, returns 429 directly, no surprise four-figure bills overnight.
Row-level isolationDepartment tree inheritanceHard cost cap
🛰️
Your PC is the AI hub; phone/IDE are its satellites
Other desktop software is a single-machine tool. Lyncore's SATCOM protocol lets phone apps, a VSCode extension, and a browser extension all connect to the same Agent — automation you build on the office PC, you can check progress from your phone at home. Supports reverse-proxy NAT traversal, so even on a business trip you can inspect the sandbox on your home PC. Translation, legal, email-proxy satellites — all of Lyncore's "extra brains", and the ecosystem keeps growing.
Three-client collaborationReverse-tunnel NAT traversalSatellite node ecosystem
🔌
MCP + A2A + SkillHub
Lyncore natively supports two major protocol standards: MCP connects external tools (browser / database / API), A2A connects other agents (sub-agent collaboration / cross-agent task delegation). SkillHub marketplace installs community skill packs in one click, compatible with OpenClaw community's 80k+ skills — you don't need to start from scratch; the ecosystem has everything. Plus evolution skills that get sharper the more you use them.
MCP protocolSkillHub marketCommunity ecosystem
Continuous evolution
Others' agents die after setup, Lyncore keeps learning
Not by talk, but by the evolution loop running in code. Three stages → fully auditable end to end.
🔍
Stage 1: Discover & Propose (automatic)
| What it does | How you know |
|---|---|
| Shadow mode silently watches your repeated actions | State machine records every observation event |
| A recurring action pattern → auto-drafts a skill | You get a "skill awaiting review" notification |
| Not a black box — every step is logged | Full trace visible in the console |
🛡️
Stage 2: Shadow & Verify (safe)
| What it does | How you know |
|---|---|
| Shadow-mode dry run, preview the effect | You see "what would happen if it ran" in the sandbox |
| Sandbox guardrails: no overreach, no touching real data | Pass / fail check is clear at a glance |
| All operations stay off your real data | Runs in an isolated environment, can't reach your real files |
✅
Stage 3: Review & Activate (you decide)
| What it does | How you know |
|---|---|
| Preview the effect; approve if it looks good | One button activates it as a real skill |
| Reject if not — it keeps improving | Rejection reason feeds the next proposal |
| Every step needs your nod — won't change habits silently | Evolution audit log: who, when, what result |
It's not "auto-execute", it's "auto-learn then wait for you". Fully auditable, not a black box.
Dify workflows are hand-configured; Coze plugins are hand-written. Lyncore's Evolution Engine discovers, writes, and tests for you — you only click "Approve" at the end.
Agents that actually work
Not a chat box — your AI control console
Browser automation + desktop control + DAG workflows — three-in-one in a single window.
📊
Scenario 1: Search → clean → build knowledge base → report (end-to-end)
| You say one sentence | Lyncore does it for you |
|---|---|
| "Research the latest pricing of three competing products" | Auto-search → open 10 tabs → scrape key tables → compile a comparison report |
| "Clean this CSV and draw a trend chart" | Drop it in → auto-detect fields → clean outliers → generate visualizations |
| "Throw all contract clauses into the KB; answer anything later" | Vectorize docs → agent auto-loads → Q&A with source citations |
👉 Your job: say one sentence, drop one file. Lyncore's job: search, scrape, crunch, chart, store.
🤖
Scenario 2: Assemble agent teams, automate the desktop
| You say one sentence | Lyncore does it for you |
|---|---|
| "Write me a WeChat article with a cover image" | Set persona → bind KB → draft → image → watermark → one-click multi-platform publish |
| "Every Monday 9am, push last week's data to my WeCom" | Scheduled check → auto-pull data → push to DingTalk/Feishu/Slack |
| "Batch-process these 50 invoice screenshots for me" | AI reads UI → clicks and types → processes one by one → flags anomalies |
👉 You walk away, it works. Agent teams on call — 10x more reliable than an intern.
📅
Scenario 3: Manage schedule, email, and team
| You say one sentence | Lyncore does it for you |
|---|---|
| "Schedule a 3pm meeting with Mr. Zhang tomorrow, send the invite" | Dictate and it sends → calendar auto-blocks → attendees synced |
| "What's on my to-do today?" | Pushes today's agenda on schedule → more punctual than a secretary |
| "@Legal Agent, review this disclaimer clause for me" | Multi-agent collaboration → each owns its domain → consolidated output |
👉 It knows when you meet, who should join, what you talked about last time — no context to repeat.
Two editions, covered
Two editions: one codebase, from personal to enterprise
Personal edition is open for download; Enterprise is open for pre-orders.
Personal = all core AI power unlocked. Enterprise = multi-team collaboration + compliance audit + production-grade stability. Same codebase underneath — upgrading needs no relearning.
💡 Not sure which to pick? Upgrade from Personal to Enterprise anytime — data migrates, no re-setup needed.
⬇ Download LiteBook Enterprise demo · coming soon
Do the math
Lyncore lets you spend with clarity
Side by side: same AI capability — why does Lyncore save you enough to buy a new computer?
💰 A real calculation (assume 1M tokens called per month):
Figures based on DeepSeek's official pricing (bring-your-own-key case); for magnitude reference only. Built-in official models are billed in compute points with a platform markup — see Billing.
Lyncore saves you ¥3,500+/year — enough for a new computer. And your data stays on your own disk.
Cloud platforms take a toll between you and the model vendor; Lyncore calls the model directly with your own key — the money goes only to the vendor, through no middleman.
Save here, spend there
Lyncore doesn't lock you to one model — pick what fits
🔑
Use your own key
OpenAI, DeepSeek, Qwen, GLM — paste in the key, Lyncore never records or proxies it. The tokens you pay settle directly between you and the vendor.
No middlemanKey not stored
🦙
Ollama, zero token cost
Connect Ollama local models — works offline, swap Llama/Qwen/Gemma anytime. Compute yours, data yours, zero token fees.
Fully offline0 extra cost
💸
Route by task, save tokens
Lyncore's built-in model router auto-picks the best model by task weight. Simple tasks run Flash (fast + cheap), complex tasks run Pro (strong + accurate). Same model calls — but Lyncore cuts 70% of token spend for you.
Smart routingSaves 70% tokens
Built-in model ready out of the box
One exe, everything inside. Built-in model, and you can still plug in your own key (zero markup). Python + Node.js runtime baked in — no dependency hassle, no environment setup. Double-click, open, ship.
Personal ¥12/month · macOS / Linux coming soon
⬇ Download Lyncore Lite for Windows~120MB installer
One-click download → install → start chatting