Orchestrating agentic context by giving LLMs safe, structured access to the filesystem.
Agents are smart but forgetful. Working across multiple projects (Context A -> Context B) meant manually copying context. There was no "Clipboard Manager" for AI memory.
You find an insight in a PDF report (Agent A) but want to apply it to your code (Agent B). Moving this insight is manual friction.
Agents have no persistence. Once you close the session, the knowledge is gone.
Elevator Pitch:
A Model Context Protocol (MCP) tool allowing agents to
Read/Write/Zip/Search and "Teleport" context between sessions via a persistent SQLite
memory.
| Capability | Traditional Context | FileOps (MCP) |
|---|---|---|
| Persistence | Session only. | SQLite Backed Meta-Memory |
| Cross-Project | Impossible. | "Quick Save" / "Quick Load" |
A Node.js server exposing JSON-RPC endpoints to the Agent.
// The "Teleport" Logic
/*
User: "FileOps, save this TSLA report comparison."
Agent -> FileOps.save("tsla_report_q3", content);
... switch to Cursor IDE ...
User: "FileOps, load the TSLA report."
Agent -> FileOps.load("tsla_report_q3");
> Returns structured summary from SQLite
*/
Faster context switching.
Granular file access.
Proving ground for MCPs.
(Diagram archived)
Async I/O for filesystem ops.
AST manipulation.
Tool calling interface.
Superseded by MCP servers, but foundational for my understanding of agent inputs.
Back to Playground