Post-module-migration: test memory/inference, verify backwards compatibility, expand sync connectors

[Al629176]

Summary

Two major PRs have migrated large portions of the codebase into dedicated modules — provider/inference code into tinyagents (PR #4784, merged) and memory engine code into tinycortex (PR #4794, pending merge). Once PR #4794 is merged, three follow-up tasks are required before the migration can be considered complete.

Problem

Module extraction changes the internal boundaries of the codebase. Without explicit validation:

• Memory and inference behaviour may silently regress
• Existing user memory folders (from prior versions) may break on upgrade
• The new modular architecture has no sync connectors for external AI coding tools (Claude Code, Codex, etc.)

Tasks (unblock after PR #4794 merges)

1. Test memory and inference end-to-end

Verify that memory reads, writes, retrieval, and inference calls all work correctly now that the underlying code lives in tinycortex and tinyagents rather than inline in openhuman.

• Memory sync, chunking, and retrieval produce correct results
• Inference (chat, embeddings) routes correctly through the new module boundaries
• No regressions vs. pre-migration behaviour on macOS, Windows, Linux

2. Backwards compatibility — existing memory folders must survive upgrade

Users upgrading from a pre-migration build must not lose their memory data or have their existing memory folders broken.

• Existing .openhuman/users/<id>/memory_tree folder structure loads correctly after upgrade
• No data migration step required from the user
• Upgrade path tested with a real pre-migration vault (at least one test user / QA device)
• If a migration step IS required, it runs silently and automatically on first launch

3. Expand memory engine — write sync modules for external connectors

Now that the memory engine lives in tinycortex as a standalone module, build sync adapters so external tools can read/write into it.

• Sync module for Claude Code (read context from memory, write session summaries back)
• Sync module for Codex (OpenAI Codex / Cursor / Windsurf compatible)
• Adapter interface documented so additional connectors can be added without touching core
• At least one connector live and manually verified end-to-end

Acceptance criteria

• All three task groups above fully checked off
• No user-visible regression in memory or inference on a clean install or upgrade path
• At least two external sync connectors (Claude Code + one other) working against the new tinycortex module

Related

• PR feat(inference): Phase 3 P3-B — crate-native turn path + delete compatible*.rs [WIP] #4784 — provider/inference migration into tinyagents (merged)
• PR refactor(memory): complete TinyCortex engine migration #4794 — memory engine migration into tinycortex (pending merge — unblocks this issue)
• tinycortex repo: tinyhumansai/tinycortex
• tinyagents repo: tinyhumansai/tinyagents