Proposal: Agent-agnostic abstraction layer for claude-mem

[rynfar]

I recently ported claude-mem to work with OpenCode (https://github.com/opencode-ai/opencode) and it's working great. The architecture is really clean. I would prefer to contribute directly to claude-mem though instead of maintaining my own fork. But I needed something to work right away.

While doing the port, I noticed most of the Claude-specific coupling is at the edges (hook I/O format, transcript parsing, paths, exit codes) while the core (worker HTTP API, database schema, settings system) is already generic.
Proposal: Add an adapter layer that would let claude-mem work with any agent supporting hooks:
Agent Hooks → Adapter Layer → Generic Worker (unchanged)
Four interfaces would cover it:

• HookAdapter - event names, I/O format, exit codes
• TranscriptAdapter - message parsing
• ConfigAdapter - paths, defaults
• ToolFilterAdapter - skip rules
  Benefits:
• No breaking changes for existing Claude users
• Opens the door to Cursor, Aider, OpenCode, etc.
• Community can contribute adapters
• Positions claude-mem as the persistent memory solution for AI coding agents
  I've already got OpenCode working, so I could contribute that adapter as a reference implementation alongside the Claude one.

[thedotmack]

We do need a good solution to extending to other platforms, and definitely want it to be in the main branch :) @rynfar did you join the discord? Message me directly :)

[rynfar]

I couldn't find your discord listed anywhere. Can you post it?

[thedotmack]

Want to submit a PR? I am very open to this and the sooner the better :)

[rynfar]

I'll join discord and message you once i get a minute! I'll start working on a pull request for this! Excited to contribute!

[thedotmack]

Thanks man! :) Looking forward to it