pi-aiand

The provider extension that adds open-weight models to pi

One pi install and your agent gains 8 ai& models — including Kimi K3 with a 1M-token context, all wired into /model. No config, no new SDK.

30-second start

pi install npm:pi-aiand           # 1. install
export AIAND_API_KEY="sk-..."     # 2. add your key  → console.aiand.com
pi --model qwen/qwen3.6-27b "explain this repo"   # 3. code for free

$ pi --model qwen/qwen3.6-27b "add a --json flag to the CLI"

  ● reading  src/cli.ts
  ● editing  src/cli.ts   +14 −2
  ● running  bun test     ✓ 23 passed

  Added `--json`: parses argv, prints structured output, keeps the
  human-readable path as the default. Tests green.

  tokens: 4,812 in · 631 out · cost: reported by ai& in JPY

Prices below are in Japanese yen per million tokens and reflect the ai& organization used to sync this plugin. Your account's prices and model access may differ.

What this extension does

• Registers aiand as a pi provider — Qwen · DeepSeek · Kimi · GLM · Gemma · GPT-OSS show up in /model alongside your other providers.
• Zero config. No provider block to hand-write, no base URL to look up. pi install and you're done.
• Handles auth for you — env var, CLI flag, or pi --login aiand.
• Includes the current model set — model availability is synced from ai& and may vary by organization.

Models

| Model | Context | Vision | Input / 1M | Output / 1M | |-------|:-------:|:------:|:----------:|:-----------:| | qwen/qwen3.6-27b | 262K | — | ¥50 (~$0.31) | ¥510 (~$3.19) | | deepseek-ai/deepseek-v4-flash | 1M | — | ¥25 (~$0.16) | ¥40 (~$0.25) | | google/gemma-4-31b-it | 262K | ✅ | ¥30 (~$0.19) | ¥80 (~$0.50) | | openai/gpt-oss-120b | 131K | — | ¥25 (~$0.16) | ¥95 (~$0.59) | | deepseek-ai/deepseek-v4-pro | 1M | — | ¥160 (~$1.00) | ¥400 (~$2.50) | | moonshotai/kimi-k2.7-code | 262K | ✅ | ¥125 (~$0.78) | ¥560 (~$3.50) | | moonshotai/kimi-k3 | 1M | ✅ | ¥480 (~$3.00) | ¥2,000 (~$12.50) | | zai-org/glm-5.2 | 1M | — | ¥160 (~$1.00) | ¥650 (~$4.06) |

JPY per million tokens, with approximate USD equivalents at ¥160/USD. Cached input is billed at a lower rate where available. Synced from an authenticated GET /v1/models response; the ai& model catalog is organization-scoped.

Usage

pi                                          # interactive → /model to pick any ai& model
pi --model deepseek-ai/deepseek-v4-flash "refactor this to async/await"
pi --model moonshotai/kimi-k3 "review this architecture"
cat src/utils.ts | pi --model openai/gpt-oss-120b "review for bugs"

Authentication

Set an environment variable and you're done:

export AIAND_API_KEY="sk-your-api-key-here"

CLI flag

pi --api-key sk-... --model deepseek-ai/deepseek-v4-flash "your prompt"

Interactive login

pi --login aiand    # paste your key when prompted

Manual ~/.pi/agent/auth.json

{
  "aiand": { "type": "api_key", "key": "sk-your-api-key-here" }
}

Get a key at console.aiand.com.

Configuration

None needed. The extension auto-registers aiand as an openai-completions provider. The bundled model list is static for fast startup; update the package when ai& changes its model catalog.

License

MIT