OpenCode

Connect OpenCode, the terminal-native coding agent, to GreenPT Code via its OpenAI-compatible endpoint.

OpenCode is an open-source, terminal-native coding agent that supports any OpenAI-compatible endpoint through the @ai-sdk/openai-compatible package. This guide connects OpenCode to GreenPT and glm-5.2.

Prerequisites

OpenCode installed (opencode.ai/docs)
A GreenPT API key from your GreenPT dashboard

1. Set your API key

Store your key as an environment variable rather than hardcoding it in config:

# zsh
echo 'export GREENPT_API_KEY="your-greenpt-key"' >> ~/.zshrc
source ~/.zshrc

# bash
echo 'export GREENPT_API_KEY="your-greenpt-key"' >> ~/.bashrc
source ~/.bashrc

2. Add the GreenPT provider

Add a provider block to your OpenCode config. Use the global config at ~/.config/opencode/opencode.json for a user-wide setup, or a project opencode.json in your repo root if you only want GreenPT available for a specific project (project config takes precedence over global).

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "greenpt": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "GreenPT",
      "options": {
        "baseURL": "https://api.greenpt.ai/v1",
        "apiKey": "{env:GREENPT_API_KEY}"
      },
      "models": {
        "glm-5.2": {
          "name": "GLM-5.2",
          "limit": {
            "context": 1000000,
            "output": 131072
          }
        },
        "minimax-m2.5": {
          "name": "MiniMax M2.5"
        },
        "kimi-k2.6": {
          "name": "Kimi K2.6"
        }
      }
    }
  },
  "model": "greenpt/glm-5.2"
}

limit requires both context and output

If you include a limit object at all, OpenCode's config schema requires both context and output — a limit with only context set fails validation with limit.output: Missing key, and OpenCode won't load any sessions until it's fixed. For that reason limit is only set for glm-5.2 above; the other models omit it and fall back to OpenCode's defaults. Confirm each model ID matches exactly what GreenPT's /v1/models endpoint returns before publishing.

Available models

Model ID	Provider	Category	Input (€/1M)	Output (€/1M)	Context
`glm-5.2`	z.ai	Coding	€1.32	€4.62	1M
`minimax-m2.5`	MiniMax	Coding	€0.17	€0.99	200k
`kimi-k2.6`	Moonshot AI	Coding	€0.83	€3.85	256k

Notes on the config:

npm — always @ai-sdk/openai-compatible for a /v1/chat/completions-style endpoint. Only use @ai-sdk/openai if GreenPT exposes a /v1/responses-style endpoint instead.
baseURL — must point to the versioned root (ending in /v1), not the full completions path.
Model IDs are passed through unchanged. Whatever key you use under models (e.g. glm-5.2) must exactly match the model ID GreenPT's API expects — case and formatting matter.
limit is optional but recommended — it tells OpenCode how much context budget it has left during a session.

One provider block can serve multiple model families, as shown above — no need for a separate provider entry per model.

3. Select the model

Restart OpenCode, then run:

/models

GreenPT should appear as a provider with GLM-5.2 listed underneath. Select it, or set it as the default via the "model": "greenpt/glm-5.2" line shown above.

4. Verify the connection

Quick sanity check outside of OpenCode:

curl https://api.greenpt.ai/v1/models \
  -H "Authorization: Bearer $GREENPT_API_KEY"

You should get back a list including your glm-5.2 model ID. If this fails, fix the API key or endpoint before troubleshooting inside OpenCode.

One-command alternative

Instead of hand-editing the config, you can offer a connector script (following the pattern used by other GreenPT-style gateways) that writes the provider block and stores the credential automatically:

export GREENPT_API_KEY="your-greenpt-key"
npx -y @greenpt/connect --opencode

Add --project to write a project-local config instead of the global one. This is optional polish — worth building only once the manual path above is documented and stable.

Troubleshooting

Symptom	Fix
"Failed to load sessions for [project]" + `limit.output: Missing key`	You set `limit.context` on a model without also setting `limit.output`. Either add both, or remove `limit` entirely for that model
Provider doesn't show up in `/models`	Restart OpenCode after editing config; confirm the file is valid JSON/JSONC
"Model not found"	Model ID in config must exactly match the ID GreenPT's `/v1/models` endpoint returns
Authentication failed	Confirm `GREENPT_API_KEY` is set in the shell OpenCode launches from; test with the `curl` command above
Config changes not applying	Check for a competing `opencode.json` — project config overrides global config, and both override remote org defaults

Config precedence (for reference)

OpenCode merges config from multiple sources, in this order (later overrides earlier on conflicting keys only):

Remote org config (.well-known/opencode)
Global config (~/.config/opencode/opencode.json)
Custom config (OPENCODE_CONFIG env var)
Project config (opencode.json in repo root) — highest precedence

If you're rolling GreenPT out org-wide, the remote .well-known/opencode config is the place to pre-populate the provider block so individual developers don't have to.

On this page