OpenCode Evict / Published CLI · developer tooling
A CLI that reconciles OpenCode's cached model catalog with models.dev and live provider data, then exports upstream contributions.
bunx @mikthatguy/opencode-evict --dry-run
bunx @mikthatguy/opencode-evict --export-contribMerge catalog entries → refresh cache → export missing models as TOML.
01 / Problem
My model list kept falling behind provider APIs. I built a CLI that combines the local OpenCode cache, the models.dev baseline and live provider responses.
It produces a merged catalog that keeps local entries, and it can write missing models as TOML files for upstream contributions.
02 / My build
The tool locates the OS-native cache, reads the upstream catalog, requests provider data and reconciles the results. A dry run previews the operation; contribution export writes TOML for review.
Kilo is the first-class integration. Generic OpenAI-compatible endpoints are an explicit best-effort option, with bearer retry when needed.
INSTALL
$ bunx @mikthatguy/opencode-evict03 / Decisions
The tool keeps existing catalog entries, so it does not silently delete models you added by hand. I keep provider-specific discovery separate from the generic probing path.
The tool has no hosted interface. You get it through the npm package or the release binaries, with the README as the guide. Try a dry run to see how your own catalog compares.
04 / Current limits
The published CLI works for a focused workflow. The repository still needs regression tests for merging, provider responses and export control flow. I would also strengthen external payload validation, apply timeouts consistently and make cache writes atomic.
I plan to make these improvements before broadening first-class provider support.