AlphaCode + Token Harbor

An open-source terminal coding agent (alphacli.github.io, MIT license). Route it through Token Harbor to use every model with a single key.

Get a thk_… key from /dashboard/api-keys (a free account works), then:

AlphaCode accepts any OpenAI-compatible endpoint as a provider profile. Add one once. The key is read from an environment variable and is not written to disk. On macOS or Linux:

export TOKENHARBOR_API_KEY="YOUR_thk_KEY"
alphacode provider add tokenharbor-api \
  --base-url https://tokenharbor.ai/v1 \
  --model deepseek-v4.1-flash \
  --api-key-env TOKENHARBOR_API_KEY

On Windows, in PowerShell:

$env:TOKENHARBOR_API_KEY = "YOUR_thk_KEY"
alphacode provider add tokenharbor-api `
  --base-url https://tokenharbor.ai/v1 `
  --model deepseek-v4.1-flash `
  --api-key-env TOKENHARBOR_API_KEY

Then run a task with that profile:

alphacode run --provider-profile tokenharbor-api "your task"

Model names

Use a model id that Token Harbor serves. These two were tried with AlphaCode (reading a file, editing it and running a command):

• deepseek-v4.1-flash
• deepseek-v4-flash

The full list is the catalog, or GET https://tokenharbor.ai/v1/models with your key. Pass --model <id> to switch for one run.

The built-in entry

AlphaCode also ships a built-in tokenharbor provider, preset to the model tokenharbor/qwen3-max. Token Harbor does not serve that name (it answers 404 model_not_found), so a first run with the built-in entry fails until you choose a model. Add your own profile as above instead. The profile name must not be tokenharbor, which AlphaCode reserves for its built-in entry.

If you do use the built-in entry, pass a model: alphacode run -p tokenharbor -m deepseek-v4.1-flash "your task".