OpenCode API Model Changer

A lightweight local HTTPS proxy between your AI client and OpenCode.ai — automatically stripping the claude- prefix from model names so requests route correctly.

By CompileFuture

---

The Problem

OpenCode.ai's API expects model names without the claude- prefix. For example:

What your client sends	What OpenCode.ai expects
claude-sonnet-4-5	sonnet-4-5
claude-opus-4-7	opus-4-7
claude-haiku-4-5	haiku-4-5

Many AI tools (Claude Code, custom scripts, etc.) send the full claude-* model name. Without a fix, the request fails or routes to the wrong model.

---

The Solution

This proxy runs locally and sits between your client and opencode.ai/zen. Every request passes through it, and if the JSON body contains a model name starting with claude-, the proxy strips that prefix before forwarding the request.

Your Client  ──►  Local Proxy (localhost:8787)  ──►  opencode.ai/zen
                  (strips "claude-" prefix)

The proxy is completely transparent — headers, streaming SSE responses, binary payloads, and all other data pass through unmodified.

---

How It Works

Request Flow

1. Client sends POST /v1/messages  { model: "claude-sonnet-4-5", ... }
                    │
2. Proxy intercepts the request
                    │
3. Parses the JSON body
                    │
4. Detects model starts with "claude-"
   → Rewrites to "sonnet-4-5"
                    │
5. Forwards to https://opencode.ai/zen/v1/messages
   with the patched body
                    │
6. Streams the response back to the client (SSE-compatible)

Key Behaviors

• Model rewriting — Only rewrites models that start with claude-. Other model names (e.g., gpt-4o, gemini-pro) are forwarded unchanged.
• Streaming — Natively streams SSE (Server-Sent Events) responses back to your client with no buffering.
• Non-JSON payloads — Binary or form-data requests are piped through untouched.
• GET/HEAD requests — No body processing; forwarded as-is.
• Root path — Visiting / returns a plain-text response (useful for health checks).
• Error handling — If the upstream fetch fails, the proxy returns a 502 with a JSON error body.

---

Run Locally with HTTPS

The proxy runs via Wrangler's dev server with a trusted local HTTPS certificate. HTTPS is required because many AI clients refuse to send API keys over plain HTTP.

Step 1 — Clone the repo

git clone https://github.com/CompileFutureYT/OpenCode-API-Model-Changer.git
cd OpenCode-API-Model-Changer

Step 2 — Install dependencies

npm install

Step 3 — Generate a trusted local certificate

You need mkcert to create a localhost certificate that your OS trusts.

macOS

# Install mkcert
brew install mkcert nss

# Trust the local Root CA
mkcert -install

# Generate the certificate (run inside the project folder)
mkcert localhost 127.0.0.1

Windows (PowerShell as Administrator)

# Install mkcert — pick one:
winget install mkcert
# OR
choco install mkcert

# Trust the local Root CA
# (Windows will pop a security prompt — click Yes)
mkcert -install

# Generate the certificate (run inside the project folder)
mkcert localhost 127.0.0.1

This creates two files in the project folder:

File	Purpose
localhost+1.pem	The certificate
localhost+1-key.pem	The private key

These files are already listed in .gitignore and will never be committed.

Step 4 — Start the proxy

npm run dev

The proxy starts at:

https://localhost:8787

You should see Wrangler confirm it is listening on that address. Leave this terminal open while you use the proxy.

---

Usage

Point your AI client's base URL at the local proxy instead of directly at OpenCode.ai.

Claude Code (~/.claude.json or settings):

API Base URL: https://localhost:8787/v1
API Key:      your-opencode-api-key

Example with an OpenAI-compatible Python client:

import openai

client = openai.OpenAI(
    base_url="https://localhost:8787/v1",
    api_key="your-opencode-api-key",
)

response = client.chat.completions.create(
    model="claude-sonnet-4-5",   # Proxy strips "claude-" before forwarding
    messages=[{"role": "user", "content": "Hello!"}]
)

---

Credits

Built by CompileFuture. Subscribe for more AI tooling tutorials.