tanstack-ai-opencode-console

Community TanStack AI text adapter for the OpenCode v2 Console inference APIs. It calls https://opencode.ai/inference directly; it does not run the OpenCode CLI.

npm install @tanstack/ai tanstack-ai-opencode-console

import { chat } from "@tanstack/ai";
import { opencodeConsoleText } from "tanstack-ai-opencode-console";

const stream = chat({
  adapter: opencodeConsoleText("gpt-5.5"),
  messages: [{ role: "user", content: "Explain SSE in one sentence." }],
  modelOptions: { max_output_tokens: 256 },
});

for await (const event of stream) {
  if (event.type === "TEXT_MESSAGE_CONTENT") process.stdout.write(event.delta);
}

Set OPENCODE_API_KEY on the server or pass apiKey. Alternatively, use the server-only device sign-in helpers:

import {
  requestOpenCodeConsoleDeviceCode,
  completeOpenCodeConsoleDeviceSignIn,
  createOpenCodeConsoleSessionAuth,
} from "tanstack-ai-opencode-console/auth";
const device = await requestOpenCodeConsoleDeviceCode();
// Display only verificationUrl and userCode; keep deviceCode on the server.
const session = await completeOpenCodeConsoleDeviceSignIn(device);
const sessionAuth = createOpenCodeConsoleSessionAuth(session);
const adapter = opencodeConsoleText("gpt-5-nano", { session: sessionAuth });

Device sign-in requests workspace-scoped access. Console currently labels this package's own client ID "Unknown client". Token-bound workspaces cannot be switched locally. Account-wide sessions with multiple workspaces need sessionAuth.selectOrganization(id) before inference. A static session: {accessToken,orgId} or cancellable supplier also works. Session mode ignores the environment key and cannot be combined with an explicit apiKey.

The /auth entry also exports refreshOpenCodeConsoleSession and revokeOpenCodeConsoleSession. The supplier keeps rotated tokens in memory with per-session single-flight refresh. getSession() returns a copy for caller-managed secure storage. clear() deletes only local state; remote revocation is explicit. Do not send tokens to a browser. See authentication verification for evidence and limits.

Set apiKey: "" to omit authentication; gateway eligibility rules still apply. A live external free-tier call on October 5, 2026 returned HTTP 403 restricting usage to OpenCode clients. Use an eligible paid model. Requires Node 22.19+ and @tanstack/ai 0.65.x.

Exports opencodeConsoleText, OpenCodeConsoleTextAdapter, createOpenCodeConsole, listOpenCodeConsoleModels, parseOpenCodeConsoleModels, resolveOpenCodeConsoleApi, and OPENCODE_CONSOLE_BASE_URL. Import the browser-safe opencodeConsoleByok from tanstack-ai-opencode-console/byok.

The factory chooses Responses, Chat Completions, Anthropic Messages, or Gemini by model ID and accepts an explicit api override. Jev decision models are excluded. Discovery reports availability and inferred routing; it does not guarantee tools, multimodal input, or structured output. Input types default to text; declare verified inputModalities in the adapter config. Provider-specific parameters go in modelOptions.

Configuration supports a root baseURL, fetch, defaultHeaders, timeout (600000 ms), maxRetries (0), and apiKey or session authentication. Every protocol uses Console Bearer auth; sessions also send matching x-org-id and x-opencode-org-id headers. Persistent secret storage stays with the caller. Responses default to store:false; Messages defaults to max_tokens:4096. Cancellation, reasoning, tool calls, usage, and schema output delegate to the native TanStack adapters. Provider file handles are not exposed.

Go subscription inference can be configured with baseURL: "https://opencode.ai/inference/go", a stable x-opencode-session, your own User-Agent, and an explicit api where Go routing differs (MiniMax and Qwen use Messages). See Go configuration and verification for credential requirements and live evidence.

See full documentation and example app, the v2 inference reference, and the TanStack compatible adapter.

Unofficial community project. MIT license.