Why it matters
This release provides AI builders with a modular and composable set of tools to extend the capabilities of the Pi AI agent runtime. Developers can now integrate a wide range of functionalities, from enterprise workspace automation to advanced memory management and image generation, directly into their agent applications.

What changed

The Pi AI agent runtime, developed by TGYD-helige, has introduced a collection of pluggable extension packages designed to enhance agent functionality. These packages expose LLM-callable tools through the MCP (Message Communication Protocol), enabling agents to interact with a diverse set of services and capabilities. The core runtime provides essential types and persistence adapters, while the extension packages offer specialized features.

Key extension categories and their functionalities include:

Workspace Integrations

  • IM Integrations: Packages for Lark (@amaster.ai/pi-lark), WeCom (@amaster.ai/pi-wecom), and DingTalk (@amaster.ai/pi-dingtalk) provide deep integration with these enterprise communication platforms, covering calendars, documents, tasks, and chat functionalities.
  • Teamwork: The @amaster.ai/pi-teamwork package facilitates team collaboration and issue management via Multica.

Automation and Access

  • Browser Automation: @amaster.ai/pi-browser-use enables browser automation using chrome-devtools-mcp, with options for live-session retention.
  • Computer Automation: @amaster.ai/pi-computer-use offers cross-platform tools for desktop automation.
  • Web Access: @amaster.ai/pi-web-access provides web search capabilities (including Cloudflare integration), URL content extraction, and image search across configurable providers.

Memory and Task Management

  • Persistent Memory: @amaster.ai/pi-memory allows for persistent, curated memory injection into the system prompt, frozen between compactions. @amaster.ai/pi-memory-mem0 is also available.
  • Task Scheduling: @amaster.ai/pi-task-scheduler implements cron-based scheduled task management with LLM-callable tools.
  • Goal Management: @amaster.ai/pi-goal derives and pursues agent goals until conditions are met, with backstops for iteration and token usage.

Multimedia and Telemetry

  • Image Generation: @amaster.ai/pi-image-gen supports image generation via multiple providers, including OpenAI, Google Nano Banana, Alibaba Qwen-Image, and OpenRouter.
  • Video Generation: @amaster.ai/pi-video-gen handles direct video-model clips, batch rendering, and local concatenation, with support for FFmpeg and editable compositions.
  • Telemetry: @amaster.ai/pi-telemetry offers runtime tracing, TTFT, token/cost accounting, and diagnostics with exporters for Langfuse, Arize Phoenix, and OpenTelemetry.

Security

  • Security Policies: @amaster.ai/pi-security provides a resource-aware security policy engine and tool authorization.

Core packages, such as @amaster.ai/pi-shared for runtime types and @amaster.ai/pi-storage for persistence adapters (JSON-file, MySQL/Prisma), are foundational. All packages are ESM-only and published under the @amaster.ai npm scope, requiring Node.js >=24 and pnpm 10.18.3.

Why it matters for builders

This release empowers AI builders by offering a highly modular and extensible framework for creating sophisticated AI agents. The availability of specialized, LLM-callable tools means developers can rapidly assemble agents with complex functionalities without building everything from scratch. The composable nature of the packages allows for fine-grained control over agent capabilities, enabling tailored solutions for diverse applications, from enterprise automation to creative content generation.

Practical impact

Developers can now integrate advanced features like enterprise IM bots, automated web browsing, persistent agent memory, and image/video generation directly into their Pi-based AI agents. By installing specific @amaster.ai packages via pnpm, builders can extend their agent's toolset. For instance, integrating @amaster.ai/pi-lark allows an agent to interact with Lark workspaces, while @amaster.ai/pi-image-gen enables visual content creation. The runtime's design encourages host applications to manage routing, authentication, and model setup, promoting a clean separation of concerns.

Caveats and source limits

While the repository details a comprehensive set of packages and their functionalities, specific usage examples for each extension are primarily found within individual package READMEs, which are not fully detailed in the provided source. The source mentions a 'fresh release' and lists a latest version as v0.1.38, indicating this is an actively developed project, but does not provide specific release dates for individual packages or the overall suite. Independent benchmarks or performance metrics for these extensions are not available in the provided excerpts. The compatibility with specific LLM models or host environments beyond the general mention of DeepSeek Harness is also not elaborated upon.

Sources

Written with AI assistance from the linked sources; every claim below was checked against them automatically. How we produce articles.

Claim check: 7/7 supported claims - 7 evidence links - 100% avg confidence
  • The Pi AI agent runtime offers pluggable extension packages that expose LLM-callable tools.supported - github.com
  • Extension packages cover IM workspace integrations (Lark, DingTalk, WeCom), browser and computer automation, persistent memory, telemetry, task scheduling, image generation, web access, and security policies.supported - github.com
  • The Pi runtime packages are intentionally small and composable.supported - github.com
  • Host applications provide HTTP routing, authentication, model runtime setup, deployment configuration, and product-specific UI for Pi runtime applications.supported - github.com
  • All packages are ESM-only and published under the `@amaster.ai` npm scope.supported - github.com
  • The Pi AI agent runtime requires Node.js `>=24` and pnpm `10.18.3`.supported - github.com
  • The latest release version is v0.1.38.supported - github.com

Caveats

  • Single-source caution: verify critical details at the linked source.
Radar score 87/100 - how it was calculated
Reliability82
Freshness92
Novelty77
Technical89
Developer96
Ecosystem72
Confidence96
  • Reliability 82: GitHub metadata supports source trust
  • Freshness 92: Fresh GitHub activity
  • Novelty 77: Fresh GitHub release
  • Technical 89: Repository technical metadata
  • Developer 96: Developer tooling signals
  • Ecosystem 72: Fresh GitHub release
  • Confidence 96: Claims have reliable evidence
Share
XLinkedInHacker News

Discussion

Loading comments...

Related articles

Agents - Nov 18, 2026Pathiel AI Trading Agent for HyperliquidPathiel is an AI trading agent designed for Hyperliquid, utilizing Claude to trade perpetuals via MCP. It features precise stop and target orders, with a default risk setting of 3x leverage or 1% of capital per trade, and includes desk briefs for oil and metals. The project also offers a free Claude Code plugin.Agents - Nov 19, 2026jnMetaCode Releases 328 AI Agents for Chinese MarketThe agency-agents-zh project on GitHub offers a collection of 328 plug-and-play AI expert roles, with 67 specifically designed for the Chinese market. These agents support integration with 20 AI coding tools like Claude Code and Copilot, covering 22 business departments.Agents - Nov 17, 2026Capri AI Desktop: Open-Source AI Companion for WindowsCapri AI Desktop is a free, open-source application for Windows that transforms your desktop into an interactive AI companion. It supports various AI models like Claude Code and Ollama, offering features such as voice interaction, gaming overlays, and desktop automation.Agents - Nov 17, 2026BuilderIO Agent Native: Framework for Agentic AppsBuilderIO Agent Native is an open-source TypeScript framework designed for building agentic applications. It recently released version v0.1.298, indicating active development and a focus on providing tools for AI agents.Agents - Nov 20, 2026Grajo: Self-Hosted Agentic Framework and MCP PlatformGrajo is an open-source, self-hostable agentic framework and Model Context Protocol (MCP) platform. It allows users to build and run AI agents with persistent memory, research capabilities, and P2P sharing, while also serving as an MCP platform to connect existing AI clients like Claude Code and Cursor.Agents - Nov 17, 2026Overleaf CLI (olcli) v0.17.0 Adds AI Agent IntegrationThe Overleaf CLI tool, olcli, has released version 0.17.0, introducing an MCP server for AI agents. This allows AI agents to interact with Overleaf projects, prompting users before making changes.