A self-hosted conversational AI dev studio — full loop from requirement to deployment. AI writes code, fixes bugs, generates docs; after you approve the merge it deploys to a live URL. Mobile approval and takeover anywhere. Supports self-hosting and domestic models.
More than writing code — it delivers a running product. After approval, merge triggers automatic deployment and each project gets a live URL (project.company.lazyhappy.tech). PM review, client demo, exec progress check — just share a link. Deploys are versioned and rollback-able; if it won't start, AI self-heals.
Three modes: chat side-by-side, autonomous (submit and walk away), assisted (pauses when unsure). AI handles requirement breakdown, coding, testing; a 14-stage pipeline covers the full chain from requirement to merge, with humans only at key approval gates.
Native mobile app: task list, inbox approvals, chat, and scan-to-take-over your local Claude Code. Run one command, scan the QR, and approve cards or monitor tasks from your phone — keep things moving on the go. No token fee charged by the platform.
Images, videos, slides, Word, Excel, and PDF generated in chat auto-archive to the artifact hub — preview in-browser instantly, no downloads. An enterprise dashboard shows total output, contributor rankings, and type breakdown. AI output becomes reusable team assets.
Every change is reviewed via diff before merge; release then requires build + test + smoke gates to all pass — one failure blocks the release. Code changes happen in isolated worktrees; the main branch is never touched. AI code faces more gates than human releases to reach production.
Deploy via Docker Compose on your own servers — code and data never leave your perimeter. Bring your own domestic model keys (Doubao / Kimi / GLM / Tongyi / DeepSeek); the platform never touches token costs. Meets compliance needs for finance, government, and healthcare.
Beneath the delivery loop and team collaboration lies a battle-tested autonomous dev engine
Complete closed loop from requirement discovery, solution design, coding, code review, automated testing, quality validation to code merge. Humans only intervene at critical approval checkpoints
AI autonomously scans project code and roadmaps, quantifying requirement value through 5-dimension scoring. Independent cross-review ensures no low-value work gets through
Daily absolute cap → session budget → project total → per-call budget → stage lookahead deduction. Every dollar spent where it matters most
Syntax fix · logic retry · solution redesign · human escalation — autonomously picks the optimal recovery strategy on obstacles, never banging on the same wall
Async coroutine-driven engine managing multiple projects across different tech stacks simultaneously, with task-level worktree isolation
After every merge, auto-maintains a project knowledge base (structure index, change log, lessons learned). Failure root-cause becomes globally effective engineering playbook — team pitfalls aren't repeated
Like the automatons in Liezi that could sing and dance on their own, Yanshi lets AI autonomously complete the entire software development process. It's not a code completion tool — it's a team-shared conversational dev studio, moving AI dev capability from "installed on each laptop" to "shared on the team cloud": run, manage, learn, and observe in one place.
The full chain is: requirement → development → approval & merge → gates → deploy to live. Cursor gives you a diff, Devin gives you a PR, Yanshi gives you a URL you can open. End-to-end closed loop with no breakpoints. Tasks run in isolated server-side workspaces — closing your laptop doesn't stop them; stuck points pause for a human rather than guessing; costs are attributed by project and person, clear at a glance.
For teams with hard data-security requirements, self-host on your own servers with domestic models — data never leaves your perimeter. Already using Claude Code? Don't replace it — run one command, scan to take over approval and chat from your phone, free relay via the platform. Underneath: a 14-stage autonomous pipeline, 5-layer budget protection, and 4-tier smart replanning keep autonomous operation safe and controlled.
Try it free with your own project — running a few tasks beats reading any article
Try Free / See Pricing Self-Host Inquiry