Wrapper that starts glm-acp-agent authenticated with the OAuth access token the ZCode desktop app stores under ~/.zcode/v2/credentials.json (decrypted at each launch), so no Z.AI API key is required. An Z_AI_API_KEY already present in the environment takes precedence; set ZCODE_CREDENTIAL_SECRET when the credentials were written with a non-default secret. If the stored access token has expired, refresh it by signing in once in the ZCode desktop app.

This package implements the Agent Client Protocol (ACP), which standardizes communication between code editors and AI coding agents. This is the 0.20.x line required by glm-acp-agent.

Standalone Agent Client Protocol (ACP) server for Zhipu AI's GLM models under the GLM Coding Plan: streaming, tool calls, mid-session model switching, image input and session persistence. Authentication uses a Z.AI API key via the Z_AI_API_KEY environment variable (or a one-time glm-acp-agent --setup credentials file under ~/.config/); coding-plan subscribers create the key at https://z.ai/manage-apikey/apikey-list.

emacs-hermes is an Emacs front-end for Hermes Agent, driven over its dashboard/TUI gateway. It provides an ERC-style chat buffer with streaming replies, slash commands, approval/clarify prompts, Markdown-rendered replies with diffs opened in diff-mode, and Kanban/session/profile/MCP/cron browsers. A running Hermes Agent dashboard/TUI gateway is required.

hermes-agent is NousResearch's AI agent. This package provides only its Python gateway -- the JSON-RPC/WebSocket server started with hermes serve (or hermes dashboard), which is what emacs-hermes and other API clients talk to. The Electron desktop app and the Node/Ink ui-tui terminal frontend are not built.

A Model Context Protocol server that integrates with Google Drive and Google Sheets, letting an MCP client create and modify spreadsheets: read and write cell ranges, batch updates, add sheets, and list spreadsheets in a Drive folder.

A Model Context Protocol server exposing a single sequentialthinking tool that scaffolds dynamic, reflective problem-solving: breaking a problem into revisable, branchable thoughts that build on, question, or backtrack from earlier ones as understanding deepens.

Terminal string styling done right.

Fast RFC 3986 URI parsing and resolution.

Another JSON Schema validator.

Fast deep equal comparison.

Converts Zod schemas to JSON Schemas.

Format validation for Ajv v7+.

Load a module from a string.

Officially supported Node.js SDK implementing the Model Context Protocol (MCP), the standard interface linking AI agents to external tools, prompts, and resources. This is a stdio-only build: dist/cjs and type declarations are dropped, and the declared dependency tree is trimmed to what the stdio transport actually imports -- see the trim-to-esm and trim-package-json phases.

Traverse JSON Schema documents.

Negotiate/SPNEGO proxy authentication for proxy-agents.

Prime Agent is an open-source CLI coding agent built around two core abstractions: the Recursive Language Model (RLM) and the Continual Harness. It uses a persistent IPython kernel as its primary tool, can spawn sub-agents dynamically, manages harness state (prompts, skills, memory) via CRUD operations, and can self-improve its own harness components based on trajectory analysis.

An HTTP/1.1 client, written from scratch for Node.js.

HTML5 `FormData` for Browsers and Node.

High-performance image processing library for native use and the web.

Create an ArrayBuffer instance from a Data URI string.

Safer Node.js Buffer API.

Blob & File implementation in Node.js, originally from node-fetch.