@polyweave/zhipu (1.0.7)

Published 2026-07-17 15:30:30 +00:00 by Dvorak

Set up this registry in your project .npmrc file:

@polyweave:registry=https://hub.kl1.tenere.ai/api/packages/Polyweave/npm/

To install the package using npm, run the following command:

npm install @polyweave/zhipu@1.0.7

or add it to the package.json file:

"@polyweave/zhipu": "1.0.7"

For more information on the npm registry, see the documentation.

@polyweave/zhipu

Zhipu AI / GLM adapter for Polyweave — chat, embeddings, function calling, 1M context, AllTools agent capability.

Synopsis

import { createZhipuAdapter } from '@polyweave/zhipu';
import { createTextFrame, createParamFrame, createFrameSink, FrameType } from '@polyweave/core';

var adapter = createZhipuAdapter({ apiKey: process.env.ZHIPU_API_KEY, model: 'glm-4-plus' });

adapter.source.pipe(createFrameSink(function(frame) {
  if (frame.type === FrameType.TEXT) console.log('Text:', frame.content.text);
  if (frame.type === FrameType.TOOL_CALL) console.log('Tool:', frame.content.name, frame.content.arguments);
  if (frame.type === FrameType.EMBEDDING) console.log('Embedding dims:', frame.content.dimensions);
}));

adapter.sink.write(createTextFrame('Write a short story about AI'));

// Embeddings
adapter.sink.write(createParamFrame('embedding_model', 'embedding-3'));
adapter.sink.write({ type: FrameType.EMBEDDING, id: 'emb1', timestamp: Date.now(), content: { text: 'Zhipu AI embeddings' } });

Install

npm install @polyweave/zhipu

Why

Zhipu AI's GLM series provides AllTools agent capability (function calling, web search, code interpreter, knowledge base retrieval in a single model), 1M token context, vision (GLM-4V), and embeddings. The adapter exposes these through standard Polyweave frame protocol with budget tracking and cost estimation, enabling zero-special-case integration into multi-provider routing pipelines.

API

`createZhipuAdapter(config)`

Creates a frame duplex adapter for Zhipu AI / GLM. Returns an IFrameDuplex with source, sink, setReady, and name.

Name	Type	Required	Default	Description
`config.apiKey`	`string`	No	`process.env.ZHIPU_API_KEY`	Zhipu API bearer token
`config.baseURL`	`string`	No	`https://open.bigmodel.cn/api/paas/v4`	API base URL override
`config.model`	`string`	No	`glm-4-plus`	Default chat model (overridable via PARAM)
`config.budget`	`number`	No	`0`	Hard cap in USD. When cumulative spend reaches this, incoming TEXT/IMAGE/EMBEDDING frames are rejected with `ERROR(BUDGET_EXCEEDED)` + `END`
`config.name`	`string`	No	`null`	Logical name for the node, exposed on the duplex
`config.timeout`	`number`	No	`120000`	HTTP request timeout in milliseconds

Return value: IFrameDuplex — a bidirectional frame stream.

Properties on the returned duplex:

.source — readable frame source (output frames from the adapter)
.sink — writable frame sink (input frames into the adapter)
.setReady(key) — provide the API key after construction; flushes buffered frames
.name — logical name (config.name)

Frame Protocol

Input frame	Action
`TEXT`	Appends `{role:'user', content:frame.content.text}` to conversation, sends `POST /v4/chat/completions` (streaming by default)
`IMAGE`	Vision mode — converts to `image_url` content block with optional text prompt, appended to conversation as a user message with multimodal content. Requires `url` or `data`/`b64_json` content
`EMBEDDING`	Sends `POST /v4/embeddings` with `frame.content.text` and `params.embedding_model`
`TOOL_RESULT`	Appends `{role:'tool', tool_call_id, content}` to conversation
`PARAM`	Accumulates into adapter config. `key=apiKey` triggers `setReady`; `key=budget` sets the budget cap
`COST_CHECK`	Enters dry-run mode. Buffers subsequent content frames, calls `costEstimator.estimate()`, emits a `COST` frame then `END`. If `maxBudget` exceeded, emits `ERROR(BUDGET_EXCEEDED)` + `END`
`START`	Ignored
`END`	Ignored
`TOPOLOGY_SCAN`	Emits `TOPOLOGY_REPORT` describing the adapter node, then forwards the scan frame

Output frame	Trigger
`TEXT`	Chat content deltas (streaming) or full completion body (batch)
`TOOL_CALL`	Tool call chunks from stream or complete calls from batch response
`EMBEDDING`	One per embedding request; contains `vector`, `dimensions`, and `model` in content
`RECEIPT`	Cost receipt. Progressive receipts every 50 output tokens during streaming; final receipt from usage data at end. Embedding receipts use char-count estimation (`text.length / 4`)
`COST`	Estimated cost from `COST_CHECK` dry-run, containing `bounds`, `accuracy`, `estimateSource`, `forwardEstimate`, `inputBounds`, `maxOutputTokens`
`ERROR`	Errors (see Error Conditions) followed by `END`
`END`	Always emitted after each request completes (including errors)

PARAM Keys

Key	Type	Default	Description
`model`	`string`	`glm-4-plus`	Chat model ID
`embedding_model`	`string`	`embedding-3`	Embedding model for EMBEDDING frames
`max_tokens`	`number`	`1024`	Max completion tokens
`temperature`	`number`	undefined	Sampling temperature
`top_p`	`number`	undefined	Nucleus sampling parameter
`stream`	`boolean`	`true`	Enable SSE streaming
`system`	`string`	undefined	System prompt (sent as messages[0] with role=system)
`tools`	`object[]`	undefined	Tool definitions for function calling
`tool_choice`	`string`	undefined	Tool choice constraint
`response_format`	`object`	undefined	Response format (e.g. `{type: 'json_object'}`)
`budget`	`number`	`0`	Set or override budget cap at runtime
`apiKey`	`string`	undefined	Set or override API key at runtime (triggers setReady)

Error Conditions

Every error emits an ERROR frame immediately followed by an END frame with metadata.error: true.

Condition	Error Message	Error Code	When
Budget exhausted	`'Budget exceeded'`	`BUDGET_EXCEEDED`	`budgetTotal > 0 && totalSpent >= budgetTotal` — rejected at frame entry before any API call
COST_CHECK estimate exceeds maxBudget	`'Est cost exceeds budget'`	`BUDGET_EXCEEDED`	Dry-run estimate > `costCheckOptions.maxBudget` via `costEstimator.checkBudget()`
IMAGE frame missing url/data	`'IMAGE frame requires url/data'`	`ADAPTER_ERROR`	IMAGE with no `url`, `data`, or `b64_json` content
HTTP 4xx/5xx from chat API	API error message (from `response.error.message`) or `'HTTP ' + statusCode`	`ADAPTER_ERROR`	Response `statusCode >= 400`
HTTP 4xx/5xx from embeddings API	API error message (from `response.error.message`) or `'HTTP ' + statusCode`	`ADAPTER_ERROR`	Response `statusCode >= 400`
Network error (chat or embedding)	`err.message` (e.g. ECONNREFUSED)	`ADAPTER_ERROR`	`req.on('error', ...)` fires
Request timeout	`'Timeout'`	`ADAPTER_ERROR`	`req.on('timeout', ...)` fires
JSON parse failure (chat)	`'Parse: ' + err.message`	`ADAPTER_ERROR`	Chat response body is not valid JSON
JSON parse failure (embedding)	`'Parse: ' + err.message`	`ADAPTER_ERROR`	Embedding response body is not valid JSON
Stream error	`'Stream error'` or API error from SSE chunk	`ADAPTER_ERROR`	Streaming SSE `event.error` is present
Sink paused or ended	None (frame dropped)	—	`duplex.sink.paused \|\| duplex.sink.ended` — frame is silently ignored
Missing API key	Frames buffered	—	Frames enqueue in `frameBuffer` until `setReady(key)` is called

Cost Tracking

The adapter maintains a totalSpent cumulative counter. Costs are derived from ZHIPU_COSTS in src/costs.js:

Model	Input (per 1M tokens)	Output (per 1M tokens)
`glm-4-plus`	$0.05	$0.10
`glm-4-flash`	$0.01	$0.01
`glm-4-air`	$0.01	$0.01
`glm-4-alltools`	$0.05	$0.10
`glm-4-long`	$0.05	$0.10
`glm-4v`	$0.05	$0.10
`glm-4`	$0.05	$0.10
`glm-4-flashx`	$0.01	$0.01
`embedding-3`	$0.001	$0.00

Model matching is case-insensitive substring lookup (model.toLowerCase().indexOf(key) >= 0). Unknown models return zero cost.

Embedding Endpoint

EMBEDDING frames call POST /v4/embeddings with:

{ model: params.embedding_model, input: frame.content.text }

The response data[0].embedding is emitted as an EMBEDDING frame with { vector, dimensions, model } in content. Token count is estimated as Math.ceil(text.length / 4). A RECEIPT is emitted with the estimated input cost, followed by END.

Tests

npm test

Runs rigor.test.js (property-based cost verification: pricing table oracle, non-negative costs, zero-cost at zero tokens, monotonicity) and smoke.test.js (construction with defaults/config, setReady presence, PARAM accumulation).

Requirements

Node.js 22 or later (as per engines field in package.json)
ESM only ("type": "module")
Peer dependencies: @polyweave/core, @polyweave/costs

Development Dependencies

ID	Version
@polyweave/core	*
@polyweave/costs	*
@rigor/core	*

Peer Dependencies

ID	Version
@polyweave/core	*
@polyweave/costs	*
@polyweave/errors	*

polyweave zhipu glm llm chinese agent adapter

Details

npm

2026-07-17 15:30:30 +00:00

John Dvorak

SEE LICENSE IN LICENSE

latest

8.2 KiB

Assets (1)

zhipu-1.0.7.tgz 8.2 KiB

Versions (5) View all

1.0.7

2026-07-17

1.0.8

2026-07-11

1.0.5

2026-07-11

1.0.4

2026-07-10

1.0.2