ChengYoung — Your private desktop AI agent
A built-in local inference engine works even fully offline; connect your own cloud model services when you want. No sign-up, nothing uploaded — your data stays 100% on your machine.
Make export.py support both CSV and JSON
Done — the core change:
def export_rows(rows, fmt="csv"):
if fmt == "json":
return json.dumps(rows, ensure_ascii=False)
return to_csv(rows)
Local models + your own cloud
No vendor lock-in — your runtime and data stay under your control.
Capabilities
Every item maps to a real panel in the product.
Fully offline AI
Built-in llama.cpp / GGUF engine with Vulkan GPU acceleration — full functionality without a network.
WorkflowsOne-click task chains
Capture repeatable tasks as Recipes and replay them with one click, any time.
Skills & extensionsExtend on demand
Built-in skills work out of the box; bring more tools via MCP (Model Context Protocol).
SchedulerRun on a schedule
Let tasks run automatically on a plan; results land back in your sessions.
Isolated web panel
Third-party pages open in an isolated panel inside the main window: separate profile, zero injection. Agent browser control is off by default, per-site authorized with per-action confirmation.
16 UI languages
Chinese, English and 14 more — follow your system or set them manually.
Typical scenarios
Hand it a task, get a reviewable result — from the developer's terminal to everyday work.
Code Q&A and changes
Hand it a codebase — it locates issues, explains the design and proposes reviewable edits. Changes arrive as diffs; you decide what to keep.
WritingDocs and reports
Hand it material or a topic — it organizes, drafts and polishes into documents that are ready to use.
OfflineEveryday tasks without a network
Keep working without a network — the local model summarizes, translates and rewrites in airplane mode; data never leaves the machine.
AutomationBatch file and data processing
Hand it a rule — batch renames, content extraction and scheduled summaries run automatically on a plan.
BrowserWeb page tasks
Hand it a web task — point it at a page and let it read and organize information in the isolated panel (control off by default, authorized per site).
ExtensionsPlug into your toolchain
Hand it your existing tools — connect databases, search engines and internal APIs via MCP to bring the agent into your workflow.
Skills & Extensions
Human-verified third-party extensions with install and data-flow disclosures.
Tavily Web Search
Web-wide search and structured content extraction in one call.Compatibility tested
GitHub (Official)
Let the agent work your issues, pull requests and code search directly.Compatibility tested
Exa AI Search
Semantic AI search — find pages by meaning, not just keywords.Compatibility tested
MongoDB Database
Official extension: queries, aggregations and index management in-session, against your own Atlas or local instance.Compatibility tested
Download ChengYoung (Windows)
Install directly and update in-app; the local inference engine is bundled, models download on demand.
Show SHA256 checksum command
- Per-user install — no administrator rights required; in-app updates are verified with minisign signatures.
- If Microsoft Defender SmartScreen warns on first install: that is normal while a new signature builds desktop reputation, and the publisher line shows our registered company name and location (certificate fields). Verify the SHA256 above first, then choose "More info → Run anyway".
- After the first launch, search and download a local model in Settings → Models (each model's license is governed by its model page).
- Memory guide: 8 GB to start; 16 GB RAM / 8 GB VRAM is comfortable for 7B models. GPU acceleration needs a Vulkan driver.
- macOS / Linux: not available yet.
- Other products: ChengYoung Typeset (coming soon) · LicenseDesktop (independent sub-product site).
Pricing
The app is free; local models cost nothing to run, cloud models are billed by their providers.
All features · No usage limits · No telemetry
The app itself is completely free, and no account is required. Two ways to run models: local models infer on your own hardware at zero model cost, or cloud models connect to providers you configure yourself, billed at each provider's standard rates (ChengYoung takes no cut). Workflows, skills & extensions, the scheduler and the secure embedded browser are all included.
ChengYoung Typeset is still in development; its pricing and licensing will be announced before release.