Piggyback
A self-hosted AI workspace where context lives in persistent project files — letting multiple AI tools continue work without the user restating everything.
The Problem
AI tools are mostly siloed. You can plan something in one tool, continue it in another, and revisit it later — but in practice that usually means re-explaining the context, copy-pasting summaries, and losing track of decisions. Project state is trapped inside individual chats. Users become the manual glue between systems.
What It Is
Piggyback is a self-hosted project workspace for AI-assisted work. Humans interact through a browser UI; AI tools interact through an API. Both read from and write to durable, versioned project files.
It is not a chat logger or a passive memory collector. Context enters Piggyback through explicit, intentional writes — either by the user or by a connected tool with permission. Only useful, durable information is stored.
Core Model
- Workspaces — top-level spaces owned by the user (personal, work, side-projects)
- Projects — scoped context areas inside a workspace (trip-planning, product-launch, company-x-interview)
- Files — the durable state of a project, stored as markdown or plain text (project.md, status.md, handoff.md)
- Versions — every file update creates a new version, providing history, transparency, and rollback
- Agent Connections — external AI tools granted scoped access to a project
- Permissions — intentionally simple in V1: read or write
Design Principles
- User-controlled context — the user decides what enters shared context; no passive ingestion of chat logs
- Event-driven, not ambient — triggered by explicit user actions or API calls, not background monitoring
- Self-hosted first — runs in the user's own environment; all data stays inside their infrastructure
- File-based state — markdown and text files as the primary V1 state model; no vector databases in V1
- Version history as trust — every write is tracked; proposal/approval workflows are deferred, not needed for V1
Tech Stack
Python · FastAPI · PostgreSQL · Next.js · TypeScript · Docker / Docker Compose
V1 Scope
V1 is intentionally constrained. It includes: self-hosted API, Postgres-backed persistence, browser UI, workspaces, projects, files, file versions, agent connections, and project-level read/write permissions.
V1 explicitly excludes: passive chat ingestion, semantic vector retrieval, proposal approval workflows, hosted SaaS data storage, and rich media support. The goal is one coherent, working layer — not an overbuilt system.