Shelvia
Concept

Tool Compatibility

Shelvia does not replace ChatGPT, Claude, Codex, Cursor, Gemini, Perplexity, or your coding agents. It gives them reviewed project memory, structured handoffs, and a safe way to bring results back. It is tool-agnostic by design: any AI output can be captured, and specific tools get a named handoff file.

How compatibility works

Compatibility is tracked per tool and per path: capture (getting work in), handoff (sending scoped context out), and capture-back (bringing the result home). Every path that changes memory goes through Review. No tool writes trusted memory directly.

Statuses are honest. Available means the manual capture, handoff, and capture-back loop works today by copy, paste, or file. Capture beta means browser capture is in beta for that tool. Generic means Shelvia takes the tool's output as evidence without a named format. Planned means it is on the roadmap, not shipped. Live (preview) means the programmatic path is live but agent-runtime attestation is still in progress.

The compatibility matrix

Tool                 Capture          Handoff file         Status
----                 -------          ------------         ------
ChatGPT              paste, browser   context pack (.md)   Available
Claude               paste, browser   context pack (.md)   Available
Claude Code          paste, file      CLAUDE.md            Available
OpenAI Codex         paste, file      AGENTS.md            Available
Cursor               paste, file      Cursor rule (.mdc)   Available
Gemini               paste, browser   context pack (.md)   Available
Perplexity           browser, paste   research brief       Available
NotebookLM           browser          source pack          Capture beta
GitHub Copilot       issue, PR        task brief           Planned
Replit / Devin       task brief       build brief          Planned
Lovable / Bolt / v0  paste            build brief          Generic
MCP / REST / SDK      -               reviewed memory     Live (preview)

Browser capture is beta and text-only today; it covers ChatGPT, Claude, Perplexity, Gemini, and NotebookLM. Screenshot capture and browser automation are not shipped.

Named handoff files

For coding tools, Shelvia generates the context file the tool already reads, from reviewed memory only, with secrets blocked and redacted:

  • Claude Code: a CLAUDE.md with the objective, approved decisions, constraints, and boundaries.
  • Codex and other AGENTS.md readers: an AGENTS.md with the task, rules, and definition of done.
  • Cursor: a rule file (.mdc) with the same reviewed context.
  • Gemini CLI: a GEMINI.md instructional-context file.
  • GitHub Copilot: a .github/copilot-instructions.md for the repo.
  • Any other tool: a plain Markdown context pack you paste into the chat.

The JSON export is the canonical form and is lossless — nothing is dropped, and every other file can be re-generated from it. The instruction files (CLAUDE.md, AGENTS.md, GEMINI.md, Cursor, Copilot) are focused subsets: they carry the task, decisions, rules, and boundaries, and each states honestly what it leaves out (for example, the Cursor rule omits the sources section).

Bringing results back

When a tool finishes, the result returns through Capture Back. If the handoff carried a validation contract, Shelvia checks the return against it, first a coverage signal, then a followed, partial, or ignored read, and routes the result through Review. Nothing the tool produced becomes trusted memory until a human approves it.

No official partnership

Tool names describe compatibility with imported content and generated handoff files. No official integration or partnership is implied. Shelvia is not a runtime and does not run these tools.

For runnable code samples and the developer reference, see /developers. For the trust model in depth, see /security.