pi-provider-volcengine-agent-plan

简体中文

Unofficial Pi provider for Volcengine Ark Agent Plan. This community package is not affiliated with or endorsed by Volcengine.

Features

• Native Pi provider registration and /login, with API key validation that does not start inference.
• A maintained static catalog of 12 Agent Plan models, including DeepSeek V4.1 Flash (preview), GLM 5.3 Flash, and Doubao Seed 2.1 Turbo.
• Image input for 9 models; GLM 5.3 and DeepSeek V4 Flash/Pro remain text-only.
• Streaming, thinking, and tool calls, with gateway-specific thinking controls.
• OpenAI Responses routing, except Kimi K2.7 Code, which uses Chat Completions.

Unlike the minimal configuration in Volcengine's Pi guide, this extension provides login validation, reasoning metadata, reference cost estimates, and model-specific compatibility handling.

Requirements

• Node.js 20 or newer and Pi 0.86 or later.
• A dedicated Agent Plan API key. Regular Ark and Coding Plan keys are not interchangeable with it.

Install and log in

pi install npm:pi-provider-volcengine-agent-plan

Restart Pi or run /reload, then:

/login volcengine-agent-plan

Enter your Agent Plan key and subscription tier. Invalid keys are requested again; temporary validation failures allow retrying or explicitly saving without validation. Select a model with /model, or inspect the catalog:

pi --list-models volcengine-agent-plan

For automated environments:

export ARK_AGENT_PLAN_API_KEY='...'
export ARK_AGENT_PLAN_TIER='medium'

VOLCENGINE_ARK_PLAN_API_KEY is also accepted. Tiers are small, medium, large, and max; the default is medium.

Models and tiers

• Doubao Seed 2.0 Mini and Lite
• Doubao Seed 2.1 Turbo and Seed Evolving
• DeepSeek V4 Flash, V4 Pro, and V4.1 Flash (preview)
• MiniMax M3
• GLM 5.3 and GLM 5.3 Flash
• Kimi K2.7 Code and Kimi K3

Small exposes 11 models. Medium, Large, and Max expose all 12. Kimi K3 remains Medium+ according to the official personal-plan availability table; merely seeing its card in the console does not establish Small-tier access.

The retired Seed 2.0 Code/Pro, MiniMax M2.7, Kimi K2.6, and GLM 5.2 entries are no longer offered. If one was selected previously, choose its replacement with /model.

Thinking and compatibility

• DeepSeek V4.1 Flash exposes off, low, high, and max; its thinking toggle uses an explicit gateway control. Seed 2.1 Turbo also supports explicit thinking on/off.
• GLM 5.3 Flash, GLM 5.3, and Kimi K3 expose low, high, and max; they do not offer off.
• V4.1 Flash and GLM 5.3 Flash follow Pi's native-provider effort metadata; unsupported minimal, medium, and xhigh choices are hidden.
• Kimi K2.7 Code does not support disabling thinking. It retains Chat Completions routing for tool-call compatibility.
• Successful requests do not guarantee that every model implements distinct behavior for every effort level.

Cost estimates and limitations

Pi displays estimated public API resource costs, not the Agent Plan bill or AFP consumption. For the three new models, CNY reference prices are converted using a fixed approximate 7 CNY/USD; DeepSeek V4.1 Flash uses peak-period rates. Hourly cache storage is not included. Actual subscription charges and quota remain controlled by Volcengine.

The catalog is static and may lag console changes. Full context/output limits, concurrency, and quota reporting are not live-tested. Image/video/audio generation requires separate tools or MCP integration, not these chat model cards.

Keep your API key out of source control, issue reports, and chat messages.

License

MIT