--- title: "Run OpenCode on your company's models, connectors and limits · Selan" description: "selan opencode starts OpenCode on every model your company connected, with its connectors and skills and the same limits as Claude Code and Codex. Nothing goes into OpenCode's config." url: https://selan.ai/blog/2026-10-27-opencode updated: 2026-10-27T23:00:02+02:00 --- [← All posts](https://selan.ai/blog) Blog # Run OpenCode on your company's models, connectors and limits selan opencode starts OpenCode on every model your company connected, with its connectors and skills and the same limits as Claude Code and Codex. Nothing goes into OpenCode's config. 27 October 2026FeatureIn the docs: [OpenCode](https://selan.ai/docs/opencode) Some of your developers would rather work in OpenCode than in Claude Code or Codex. Until now the company's accounts did not reach it, so a developer who wanted it at work either went without or gave OpenCode a provider key of their own. A key in a config file on a laptop spends outside every limit you set, and its usage belongs to a key rather than to a person. Its prompts skip the secret scanning the rest of the team's requests go through, and the company's connectors and skills are missing from it. ![A terminal running selan help opencode. The page reads: selan opencode, launch OpenCode on your company's connection. Usage: selan opencode \[args\], and --model provider:model, a model your company serves, checked before OpenCode starts. Then: every argument goes to OpenCode as typed, and nothing is written into OpenCode's config.](https://selan.ai/img/blog/2026-10-27-opencode/help.webp) selan help opencode, from selan 1.24.0. Selan 1.24.0 adds `selan opencode`. It starts OpenCode on your company's connection, with the models, connectors and skills your company set up, and every request it makes goes through the same rules as Claude Code and Codex. ## Run it ```bash selan update brew install opencode selan opencode ``` It needs OpenCode 2. Homebrew installs it, and so does OpenCode's own installer, `curl -fsSL https://opencode.ai/v2/install | bash`. On Windows, run `scoop install opencode`. OpenCode opens on a provider called Selan. Every argument passes through to OpenCode, so `selan opencode run "fix the tests"` runs one prompt. To start on a particular model, name it with its provider: ```bash selan opencode --model vertex:gemini-3.8-flash ``` Selan checks the model with your company before OpenCode starts, and refuses one your company does not serve, with the reason. OpenCode's own `-m selan/` flag and its `/models` picker work too. ## What OpenCode gets Every model in your company's Selan catalog is there, with its name and context size, read from Selan's gateway at each launch: Claude, OpenAI's GPT, Gemini on Vertex, OpenRouter, Ollama, whatever your company connected. Only Selan's provider is enabled for that launch. Your company's MCP connectors come with it. The ones Selan hosts, such as Brave Search, Slack and Selan, work at once on your session. One that needs its own sign-in asks for it through OpenCode the first time you use it. If your own OpenCode config already defines a connector with the same name, your server is kept. So do your company's plugin skills. Its plugin marketplaces are fetched with git into your selan directory, and their skills are offered to OpenCode. ## Nothing lands in OpenCode's config Selan writes nothing into OpenCode's config and leaves no credential on disk. Each launch hands OpenCode a private server and its configuration in the environment, so a bare `opencode` keeps whatever it had. ## The same rules as Claude Code and Codex Every request goes from your machine through Selan, so what your company set for Claude Code and Codex holds in OpenCode too: who is calling, which models are allowed, spend and concurrency limits, secret scanning before a provider sees the prompt, the log of database calls, and usage counted to the person who made the request. We ran OpenCode 2.2.0 through it, and its requests were served, tool calls included. The one exception is a Claude model served from a Claude subscription, which answers with Anthropic's own message that third-party apps now draw from extra usage, not plan limits. ## Why OpenCode, and not Cursor Cursor can use your own key, but in that mode every request goes from Cursor's servers, so the prompt and the key pass through Cursor before they reach Selan. Its Tab, Apply and indexing never leave Cursor at all. OpenCode runs on your machine and talks to Selan directly, so every request it makes is one your company's rules apply to. The setup, and what does not work, is on the [OpenCode docs page](https://selan.ai/docs/opencode). **How it works today.** A post says what changed on its day; the docs say how Selan works now. [OpenCode](https://selan.ai/docs/opencode)