Pige is a local-first general-purpose Agent with personal knowledge as its strongest advantage. Ask, paste, or drop something once; Pige answers now and keeps useful narrative and structured knowledge growing locally without making you supervise every step.
The same Agent decides whether to answer directly, retrieve, inspect, parse, OCR, analyze, or grow durable knowledge. Pige preserves attached sources and constrains tools, but does not replace that decision with format rules or a fixed workflow.
Your files remain sources. Open local knowledge stays yours. Markdown holds narrative knowledge; Dataset Bundles hold structured knowledge. No folder, schema, tag, title, or capture mode is required before you begin.
Pige is in active pre-alpha development. There is no public installer yet. Do not use it as the only copy of important knowledge.
Ask or drop → Agent decides → Knowledge keeps growing
Pige answers normally when local evidence is irrelevant. When material matters, Pi selects bounded tools to preserve, understand, connect, and maintain cited Markdown autonomously; operation history and undo replace routine approval.
Most assistants forget your working context; note apps ask you to organize too early. Pige combines a general Agent with a local knowledge compiler that can remember useful work beyond one conversation.
- Conversation when you need it, durable knowledge when it matters.
- Inspectable and portable. Sources and citations stay attached; Markdown and Dataset Bundles are durable truth, while indexes and caches are rebuildable.
- Calm outside, strong inside. Complex tools and safeguards stay behind useful defaults.
Today, the pre-alpha can:
- Capture text, web pages, Markdown, TXT, PDF, DOCX, PPTX, and images while preserving files as managed copies or verified references.
- Extract structured document content and run supported macOS Vision OCR paths.
- Run preserved text and supported PDF/DOCX/PPTX/images through embedded Pi, isolated BYOK binding, bounded inspect/parse/OCR tools, and validated knowledge publication.
- Generate source-backed Markdown and run a bounded cited Home knowledge-answer path.
- Review an exact staged create-note proposal from Home; this is a transitional exception path.
- Use persistent-job recovery and local-backup foundations.
Unified text/one-file/static-URL Provider-to-Home plus exact durable follow-up now runs. Autonomous knowledge growth and broader recovery remain pre-alpha work. See the implementation playbook.
- Complete autonomous knowledge growth: eligibility, Activity/Undo, and exceptional recovery across normal note/link/organize work.
- Finish the Agent spine: multi-source work, cross-platform document/OCR, and stronger recovery.
- Ship a trustworthy alpha: backup/restore, accessibility, localization, packaged macOS/Windows proof, signing, and updates.
Sync, mobile, and collaboration come later. See the roadmap and product requirements.
Your vault and durable truth remain local; that is a data-ownership principle, not a confirmation workflow. Pige requires no cloud account or telemetry. Once connected, a BYOK provider receives bounded selected context seamlessly—not the whole vault by default.
Built by Agents. Directed by humans.
AI Coding Agents author and maintain Pige's implementation code. Humans set direction, review evidence, and authorize releases. Contracts, tests, traceability, and CI are the project memory on the path toward safe planning-to-release automation.
Watch for the first Public Alpha, explore the vision, or help through Contributing.
For AI Coding Agents: start with repository instructions and the task router.
Security · Privacy · Support · Code of Conduct · Apache 2.0