Mike fork being rewritten as Luke: a local-first, single-user job-search workbench targeting Ollama, SurrealDB, and an eventual Go backend
CaliLuke/mike started as a standard Mike fork and is now actively rebranding as Luke, with a documented 8-milestone plan to replace the cloud stack. Supabase Auth, Postgres, R2 storage, and Express are all on the way out in favor of embedded SurrealDB/SurrealKV, a Loom-generated Go backend, Romancy durable workflows, and local file storage. Ollama is the default AI surface; gemma4 is the default model across all tiers.
The fork still ships upstream-compatible maintenance work - a shared_with encoding bugfix, inline project creation in the tabular-review modal - alongside the migration infrastructure. The M0 persistence spike (a Rust FFI bridge to SurrealKV via CGO) is already committed. A Codex Supabase MCP config is present for dev tooling.
For an outside evaluator: the maintenance commits (shared-with bug, modal UX, auth-gated reload) are selectively importable. The local-first architecture work is not - it assumes a single-user no-cloud deployment and introduces a macOS arm64-verified Rust build dependency. The LICENSE file was deleted from the repository; the README carries only a one-line AGPL reference.
brandingpersonasinfrastructure
This fork has stopped being a legal-AI tool entirely - it's being rebuilt as a single-user app that runs start-to-finish on your own machine.
One commit reframes the entire repo - new name, new purpose, new stack, eight-milestone migration plan, and a working Rust FFI bridge to embedded SurrealDB. This fork is diverging hard from upstream.
Projects you'd shared with a colleague simply weren't showing up - and nothing told anyone it had failed.
Two of three call sites passing user email to Supabase's `.contains("shared_with", ...)` were sending a raw JS array instead of a JSON-encoded string, so the `@>` predicate matched nothing. Users with shared projects saw empty lists.
A small workflow fix that removes one of those annoying detours where starting a new review means bouncing through three screens first.
Three commits landed within ~30 minutes that let a user create a new project without leaving the "Add new tabular review" modal - and then fix the race condition the change introduced.
A new option to run the underlying AI models on your own machine instead of calling out to a paid cloud service.
Adds Ollama as a third LLM provider and extends the document pipeline to handle Markdown and plain-text files. The useful parts are well-isolated - but the fork also flips every model default to gemma4 and runs a schema migration to match, which breaks any deployment without a local Ollama server.