Skip to main content
Install the Dari CLI, then launch OpenCode through your organization’s default router:
On first use, Dari signs you in if needed, creates and privately caches a Routing key, and offers to connect a ChatGPT or Claude Code subscription or use Dari managed billing. It then starts the installed opencode executable with session-local configuration. Your OpenCode config files are left intact. On first launch, setup opens an Accounts page where you can connect one or more ChatGPT or Claude Code subscriptions or choose Dari managed billing. Later launches go straight to the agent; run dari opencode --connect to open the page again. Without the Autorouting plan, OpenCode starts on your router’s first model and /models switches between them; run dari opencode --connect and choose Get Autorouting to subscribe so Dari picks for you. Every following argument is forwarded to OpenCode:
The launcher supplies its configuration through OPENCODE_CONFIG_CONTENT, replacing any value you set for that variable. It adds a Dari provider with the dari/routing model, a one-million-token context window, and image input, and pins OpenCode’s built-in agents (including Scout), subagents, and title, summary, and compaction tasks to dari/routing. OpenCode sends its session ID as the x-session-affinity header, so requests from one session group under the same conversation in Activity. Subagents appear in that conversation’s Subagents view, each with its own routing lease. The launcher’s plugin also labels title generation as Auxiliary and context compaction as Compaction, both served by the cheapest routable model. Dari chooses the model and reasoning level per request. OpenCode normally connects to a shared background service, which would not see session-local configuration. Interactive, run, mini, api, models, and stats launches therefore add --standalone to start a private server. --server and pair are rejected because they use a server Dari cannot configure; run dari opencode serve to connect other clients. Management commands such as service, session, and auth run without Dari’s configuration, so OpenCode’s long-lived background service never picks it up. --help and --version skip sign-in and onboarding. A launcher plugin shows the routing state in OpenCode’s prompt footer: the model that served the turn, its thinking level when one is enabled, how many turns remain on the current routing lease, and whether the last response read from the provider’s prompt cache. Routing values come from the X-Dari-Selected-Model, X-Dari-Thinking-Level, and X-Dari-Lease-Turns-Remaining response headers; cache status comes from response usage and expires after five minutes. Set DARI_ROUTING_API_KEY before launching to use an existing Routing key instead of the CLI-managed key. Set DARI_OPENCODE_PATH when opencode is not on PATH.

​
Manual Configuration

If you do not want to use the launcher, create a Routing API key:
Add this provider to ~/.config/opencode/opencode.json:
"reasoning": false keeps OpenCode from sending a fixed reasoning level; Dari still chooses one per request. The routing status footer is available only through the launcher. To use a different model set, edit the default router or Create A Router. For a specific router, point baseURL at https://routing.dari.dev/{router_id}/v1 and set limit.context to the smallest window among its enabled models.