No LSP
OpenCode configures language servers. Kolkrabbi does not, and the plan says so bluntly: it is a module-sized piece of work and the budget is spent.
Compare
Kolkrabbi
vs OpenCode
OpenCode is the closest thing to a sibling on this list: MIT, open source, terminal-first, model-agnostic, written in TypeScript, with 75+ providers through the AI SDK and Models.dev, configurable LSP and MCP servers, a desktop app and an IDE extension. It is also far larger — over 200,000 stars. Kolkrabbi reaches for the same ground from a different direction: one static Go binary, a dial that chooses models rather than thinking tokens, and a local record of what every model cost and how good you judged it.
Every claim below was checked against primary documentation on . Both projects move quickly. If something here is out of date, it is a bug — say so.
01 / WHERE KOLKRABBI IS BEHIND
What OpenCode does better.
This section comes first on purpose. A comparison that opens with its own strengths is an advertisement, and you should not trust the rest of it.
No MCP servers
OpenCode has them. Kolkrabbi does not yet.
Six providers against 75+
OpenCode inherits a provider catalogue through the AI SDK and Models.dev. Kolkrabbi has six direct routes plus everything behind the OpenRouter gateway — which covers most of the same models, but through one door rather than many.
The terminal, and only the terminal
OpenCode ships a desktop app for macOS, Windows and Linux and an IDE extension, and can share a session as a link. Kolkrabbi is a terminal program with no hosted service at all.
203,579 stars against a project you have not heard of
That is a real difference in bug reports, plugins, packaging and the odds that someone has already hit your problem.
02 / WHERE KOLKRABBI IS AHEAD
What Kolkrabbi does differently.
Not better at everything — different in specific ways, each of which you can check in the source or in the first ten minutes of using it.
Both subscriptions, one terminal
Kolkrabbi runs turns on your Claude plan through your own claude CLI and on your ChatGPT plan through your own Codex CLI, without ever holding a credential. OpenCode's own documentation says its Claude Pro/Max sign-in was removed in 1.3.0; it offers ChatGPT Plus, GitHub Copilot and GitLab Duo. One caveat, because it is the first thing you would hit: starting a session still expects a gateway key even when a plan answers the turns. That is on the list to remove, not a decision.
The dial picks the model
One control — low, medium, high, max — decides which model runs the turn and how wide the agent work goes. Everywhere else, choosing a cheaper model for cheap work is a separate decision you have to remember to make.
A local ledger of cost and quality
Calls, tokens, cost, latency and your own 1–5 ratings per model, in a JSONL file on your disk, summarised by /stats inside the session. This is the feature Kolkrabbi is actually built around.
One static Go binary, no runtime
Under 11 MiB, two dependencies, milliseconds to start, nothing to install underneath it. OpenCode is TypeScript and brings a runtime with it.
Nothing leaves without you asking
No telemetry, no analytics, no background version check, no hosted service and no session sharing. Sessions are files on your disk. That is a deliberate refusal, not a missing feature.
03 / SIDE BY SIDE
The whole thing, in one table.
Where a cell says a fact was not established, it means exactly that: we did not find it in the documentation we read, and we would rather leave a gap than guess.
| Capability | Kolkrabbi | OpenCode |
|---|---|---|
| License | Apache-2.0 | MIT |
| Written in | Go. One static binary under 11 MiB, two dependencies | TypeScript |
| Model providers | OpenRouter, plus Anthropic, OpenAI, Ollama, vLLM and LiteLLM directly | 75+ through the AI SDK and Models.dev |
| Local models | Yes — Ollama, vLLM, or any OpenAI-compatible URL | Yes — Ollama, LM Studio, llama.cpp, and any OpenAI-compatible base URL |
| A subscription you already pay for | Claude, through your own claude CLI, and ChatGPT, through your own Codex CLI |
ChatGPT Plus, GitHub Copilot and GitLab Duo. Their docs say Claude Pro/Max sign-in was removed in 1.3.0 |
| Which model runs the turn | The effort dial picks it: low, medium, high, max map to models you choose | You pick the model. |
| Undo | Checkpoints before every write, then /rewind for files or /undo for files and conversation |
/undo and /redo, several steps back |
| Cost and quality data | A local JSONL you own: calls, tokens, cost, latency and your own 1–5 ratings | Not documented on the pages we checked |
| MCP | Not yet | Yes |
| LSP | Not yet | Yes, configurable LSP servers |
| Subagents | Dependency-ordered tasks sharing one working tree | A build agent and a read-only plan agent on Tab, plus an @general subagent |
| Sharing a session | No hosted service. Sessions are files on your disk | /share makes a link. Off by default |
| Operating systems | macOS and Linux | macOS, Windows and Linux, with a desktop app and an IDE extension |
04 / CHOOSING
Which one you should actually use.
OpenCode is the more complete tool today, and saying otherwise would be silly. Here is the honest split.
When you want the ecosystem
You need LSP or MCP, you want an IDE extension or a desktop app, you want to share a session, or you simply want the project with more people fixing it.
When you want to know what models cost you
You are moving between cloud and local models and want the cost, latency and quality of each one recorded on your own disk. Or you pay for both Claude and ChatGPT and want one terminal that uses both. Or you want one small binary and no telemetry.
They do not conflict
Both install with one command and neither takes over your machine. Running them against the same repository for a week is a better comparison than this page is.
05 / TRY BOTH
Ten minutes settles more than a table.
Both are one command. Run them on the same repository, on the same task. When we did that with a local 8B model, all three scored zero — the pre-registered benchmark is where side-by-side results will land as they exist.
| Tool | Install |
|---|---|
| Kolkrabbi | curl -fsSL https://kolkrabbi.francomichetti.com/install.sh | sh |
| OpenCode | curl -fsSL https://opencode.ai/install | bash |
ALSO COMPARE
Kolkrabbi vs Claude Code Kolkrabbi vs Codex CLI All comparisons