DeepSeek OpenCode Setup: /connect, Version Gate, Flash

Introduction

Search DeepSeek OpenCode setup, and the blockers are often not the key — old OpenCode builds, wrong provider, or treating OpenCode like DeepSeek Harness / Codex with the same config.

Short path: upgrade OpenCode to ≥ v1.18.30, launch it, run /connect → pick deepseek → paste your API key → choose DeepSeek-V4-Flash per official docs. This site is an independent community; steps follow official Integrate with OpenCode. Models and peak pricing: pricing page.

I. OpenCode Is Not Harness / Codex / Copilot

You want	Use	On this site
Open-source coding assistant (terminal / web) with DeepSeek in the official provider list	OpenCode	This article
OpenAI Codex CLI / desktop / VS Code extension	Codex + Responses API	DeepSeek Codex setup
Stay in GitHub Copilot Chat or Copilot CLI	Copilot extension / BYOK	DeepSeek GitHub Copilot setup
DeepSeek’s own agent runtime layer	Harness	Harness getting started / v0.2.1-alpha.2

Paths can coexist, but do not copy configs across clients: OpenCode uses /connect and its own provider settings; Codex edits ~/.codex; Copilot CLI needs the anthropic provider type. Keys come from the DeepSeek Platform and bill DeepSeek API balance — free web chat is unrelated. Overview: DeepSeek API.

II. Migration: Already on OpenCode, Switch to DeepSeek

Official flow: upgrade first, then connect the provider:

1. Run opencode upgrade and confirm ≥ v1.18.30 (older builds hit compatibility issues; trust opencode --version on your machine).
2. Run opencode to launch.
3. In the input box, type /connect, search for and select provider deepseek.
4. Paste your DeepSeek API Key (usually sk- prefix).
5. Per official docs, pick DeepSeek-V4-Flash.

Flash fits daily coding and high-frequency small edits; switch to Pro in the model list for hard reasoning or long refactors when OpenCode exposes it. Thinking effort vs billing: DeepSeek thinking mode.

III. Fresh Install: Download Page + Same /connect Flow

1. Install or upgrade from the OpenCode download page; still aim for ≥ v1.18.30.
2. cd into your project and run opencode.
3. /connect → deepseek → Key → DeepSeek-V4-Flash.

Right after install, validate with a small compilable change: streaming replies and tool calls work before you trust heavy agent jobs. If you hit rate limits, see DeepSeek API 429.

IV. Optional: Pin Provider in opencode.json

For most people /connect is enough. To hard-code DeepSeek in repo or machine config (CI, multi-machine sync), use OpenCode’s provider section with an OpenAI-compatible package pointing at https://api.deepseek.com and declare model IDs. Notes:

• Provider ID in /connect must match the ID in config (usually deepseek).

• Chat Completions compatibility often uses @ai-sdk/openai-compatible; if a model needs the Responses path, swap npm packages per OpenCode docs — do not mix blindly.

• Store the key in env (e.g. DEEPSEEK_API_KEY); never commit sk- keys to Git.

Field names follow current OpenCode Providers; DeepSeek facts stay on Integrate with OpenCode.

FAQ

Why does official docs default to Flash instead of Pro?
In coding-assistant workloads, Flash (including current deepseek-flash / V4.1 Flash capability) balances latency and cost; the integration page currently says DeepSeek-V4-Flash. Reserve Pro for harder reasoning — see What is DeepSeek V4 Pro.

What if old “custom base URL proxy” tutorials conflict?
Start with official /connect provider flow. Only fall back to compatible endpoints or third-party relays when OpenCode is too old or lacks built-in DeepSeek — and accept protocol differences yourself.

Can one key serve Codex / Copilot too?
Yes — same DeepSeek account balance. Watch concurrency and billing when several heavy agent clients run together; peak windows: DeepSeek API peak hours.

Summary

Shortest DeepSeek OpenCode setup: version ≥ v1.18.30 → opencode → /connect → deepseek → Key → DeepSeek-V4-Flash. Pick the client first, then config; for Codex / Copilot / Harness, use their guides — do not paste one env everywhere.

Further reading

• DeepSeek API
• DeepSeek Codex setup
• DeepSeek GitHub Copilot setup
• DeepSeek Harness v0.2.1-alpha.2
• DeepSeek thinking mode