langchain-pi-ts

A LangChain BaseChatModel adapter for @earendil-works/pi-ai. Use Pi — and any provider, model and credential it resolves — from LangChain and LangGraph, with native tool calling and streaming.

The TypeScript half of a pair. A Python twin (langchain-pi) reuses the same Pi configuration from LangChain/LangGraph in Python.

Install

npm install langchain-pi-ts @langchain/core

@langchain/core is a peer dependency — your app owns its version (^1.1.44). The package is ESM-only and requires Node >= 22.19.0 (inherited from pi-ai and pi-coding-agent).

Usage

createChat is the simplest entry — pass a provider and it routes to the right backend (opencode/opencode-go → native Zen, claude-code → native Anthropic, everything else → Pi):

import { createChat } from "langchain-pi-ts";

const free = createChat({ provider: "opencode", model: "deepseek-v4-flash-free" }); // free, no key
const go = createChat({ provider: "opencode-go", model: "glm-5", apiKey: "..." }); // subscription
const codex = createChat({ provider: "openai-codex", model: "gpt-5.3-codex-spark" }); // via Pi
const claude = createChat({ provider: "claude-code", model: "claude-sonnet-4-6" }); // Claude Code subscription

Or construct a backend directly:

import { ChatPi } from "langchain-pi-ts";

const model = new ChatPi({
  provider: "openai-codex",
  modelId: "gpt-5.3-codex-spark",
  reasoning: "minimal",
  system: "You are a helpful assistant.",
});

const res = await model.invoke("Hello!");
console.log(res.content);

Tool calling

ChatPi accepts any LangChain tool (Zod / tool()); schemas are converted to the JSON Schema pi-ai expects.

import { tool } from "@langchain/core/tools";
import { z } from "zod";

const getWeather = tool(({ city }) => `Sunny in ${city}.`, {
  name: "get_weather",
  description: "Get the weather for a city.",
  schema: z.object({ city: z.string() }),
});

const withTools = model.bindTools([getWeather]);

Streaming

for await (const chunk of await model.stream("Write a haiku.")) {
  process.stdout.write(chunk.content as string);
}

Credentials

Models and credentials resolve through pi-coding-agent's ModelRegistry / AuthStorage. Any provider authenticated in ~/.pi works with no key. ChatPi lazily builds a shared registry on first use (importing the package touches no filesystem); pass a custom one via new ChatPi({ provider, modelId, registry }), or getDefaultAuthStorage().setRuntimeApiKey(provider, key) to inject a key.

For OpenCode Zen, prefer the native ChatOpencode below.

Native opencode/Zen (no Pi)

OpenCode Zen is an OpenAI-compatible endpoint, so you don't need Pi. The langchain-pi-ts/opencode subpath ships ChatOpencode, a thin ChatOpenAI subclass pointed at it (@langchain/openai ships as a dependency).

The API key is optional: free models work with no key (anonymous, IP-rate- limited). For paid models pass a key via OPENCODE_API_KEY (env) or apiKey.

import { ChatOpencode } from "langchain-pi-ts/opencode";

const free = new ChatOpencode({ model: "deepseek-v4-flash-free" }); // no key
const paid = new ChatOpencode({ model: "glm-5" }); // OPENCODE_API_KEY or apiKey
const go = new ChatOpencode({ model: "glm-5", tier: "go" });

Free models: deepseek-v4-flash-free, big-pickle, mimo-v2.5-free, nemotron-3-super-free.

Claude Code / Anthropic subscription (no Pi)

ChatClaudeCode talks to the Anthropic Messages API (via @anthropic-ai/sdk) authenticated with the Claude Code OAuth session already on the machine (~/.claude/.credentials.json), billing requests against your Claude Code subscription — no API key. It needs the Claude Code CLI logged in once. The langchain-pi-ts/claude-code subpath ships it.

import { ChatClaudeCode } from "langchain-pi-ts/claude-code";

const chat = new ChatClaudeCode({ model: "claude-sonnet-4-6" });
const opus = new ChatClaudeCode({ model: "claude-opus-4-8", reasoning: "medium" });
// or: createChat({ provider: "claude-code", model: "claude-opus-4-8" })

Models: claude-opus-4-8, claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5. Adaptive thinking on Opus 4.8/4.7, a token budget on Sonnet 4.6 (Haiku has no reasoning). The 1M-context beta is opt-in via longContext: true; the subscription rejects long-context requests without extra credits otherwise. Tool calling and streaming work as with ChatPi.

Using a subscription OAuth session from a third-party app may violate Anthropic's terms and risk your account — see the pi-claude-code-auth README.

Notes

• ESM + Node only. pi-ai / pi-coding-agent are ESM-only and Node-native.
• Tool-call deltas and usage/cost metadata are preserved.

License

MIT