Run OpenCode through your VelATL balance.

1-minute setup

Create a key in Settings → API keys (name it e.g. “opencode laptop” - copy the secret once, revoke anytime), then add the VelATL provider to opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "velatl": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "VelATL",
      "options": {
        "baseURL": "https://velatl.ai/api/v1",
        "apiKey": "velatl_..."
      },
      "models": {
        "fast": { "name": "Swift (gpt-oss-20b)" },
        "balanced": { "name": "Balanced (gpt-oss-120b)" }
      }
    }
  }
}
# then use model: "velatl/fast" or "velatl/balanced"

curl https://velatl.ai/api/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer velatl_..." \
  -d '{"model":"balanced","messages":[{"role":"user","content":"Hi"}]}'
# 200 + assistant text = your key and baseURL are correct.

What works

• Streaming, tool calling and JSON mode on both protocols (text + tools only).
• Model values: balanced | fast - same slugs as the workspace picker.
• tool_choice must be auto/none; the legacy functions API is rejected with a clear 400 before billing.
• Out of credits → 402 insufficient_credits - top up in billing, no silent failure.
• Every response carries X-Request-Id - quote it in support mail.

Control agent spend

Coding agents consume far more than casual chat: long context, tool loops and retries all bill as input tokens, including your tool schemas. Three guardrails keep this predictable:

• One key per machine. Name keys per laptop or project so Settings shows exactly where spend went.
• Per-key spend caps. New keys start capped at 5,000 credits/month, counting settled and in-flight spend; raise per project or lift it in Settings.
• Keep tool lists lean. Every tool definition bills as input tokens on every call - expose only what the task needs.

curl https://velatl.ai/api/v1/models -H "Authorization: Bearer velatl_..."
# per-1k credit rates included; failed calls refund automatically.

Why VelATL for OpenCode

Honest alternative: pure API users who want 200 models should compare VelATL vs OpenRouter. Full compatibility list lives in the developer docs.