SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentsshire

shire

Shire is an infrastructure layer that lets AI agents run persistently and collaborate. Unlike the common "issue instruction—execute—output" model, agents in Shire persist across sessions, can autonomously discover each other and collaborate, continue building on yesterday's work, and users can provide feedback, iterate, and adjust direction just as if working with a real team. It does not lock users into a single model provider and supports agent harnesses such as Claude Code, OpenCode, Pi Agent, and Codex. Main capabilities include: autonomous agent communication and direct messaging, shared context; a community-maintained catalog of prebuilt agent templates; a shared filesystem used by all agents; one-time or recurring scheduled tasks; alert channels configurable to Discord, Slack, or Telegram; custom skills persisted per agent; multi-project architecture; a real-time dashboard with streaming updates; and optional username/password authentication. Shire is installed globally via npm with no other dependencies, stores data in ~/.shire/, uses SQLite by default, and runs agents as local processes with no external services required.

Open website

About this tool

Purpose

Shire is positioned as "the place where agents live and thrive," an infrastructure layer for AI agents. Most AI agent tools follow the same pattern: the user gives an instruction, the agent executes it, returns output, and then the agent disappears, starting from scratch next time. Shire tries to change this—letting agents persist across sessions, communicate autonomously with each other, and continue building on existing work. Users provide feedback, iterate, and adjust direction, in an experience closer to collaborating with a real team.

Main capabilities

  • Compatible with any model: Not locked into a single AI provider, and supports agent harnesses such as Claude Code, OpenCode, Pi Agent, and Codex. As an infrastructure layer, Shire lets users choose the model that suits their workflow.
  • Autonomous agent communication: Agents can discover peers on their own and collaborate without an orchestrator. Direct messaging and shared context enable real team collaboration among agents.
  • Community catalog: Browse and deploy a community-maintained library of prebuilt agent templates (powered by agency-agents) to quickly launch a capable team.
  • Shared drive: A common filesystem shared by all agents, making it easy to collaborate around shared artifacts.
  • Scheduled tasks: Automate agent work through one-time or recurring scheduled messages, with custom intervals, so agents can run automatically.
  • Alert channels: Get notified when agents need attention, with per-project configuration to send alerts to Discord, Slack, or Telegram.
  • Agent skills: Upload custom tools and instructions to extend agent capabilities, with skills persisted per agent across sessions.
  • Multi-project architecture: Organize agents into different projects, each with its own shared drive, settings, and environment.
  • Real-time dashboard: Monitor, chat with, and manage agents through a real-time Web UI with streaming updates.
  • Optional authentication: Protect the instance with username/password login, based on JWT, with automatic token refresh, rate limiting, and secure cookie handling; disabled by default for local use.

How to use

Install Shire globally via npm with no other dependencies:

npm install -g agents-shire

After installation, visit localhost:8080 to open the dashboard. This command installs a standalone binary for the corresponding platform (macOS, Linux, or Windows), and all data is stored in ~/.shire/.

Configuring the Agent Runtime Framework

Shire supports multiple agent runtime frameworks, and each agent's corresponding framework type is configured in the dashboard.

Claude Code: Install Claude Code and complete authentication using one of the following methods.

OpenCode: Install OpenCode and set the API key for the provider you use. OpenCode automatically detects API keys from environment variables and .env files; Shire automatically manages the OpenCode server process, so there is no need to manually run opencode serve. Note the model format: OpenCode requires the provider/model format (for example anthropic/claude-sonnet-4-6, openrouter/google/gemini-2.5-pro), which differs from other frameworks that only accept model names. For the full list of supported providers and models, see the OpenCode provider documentation.

Pi Agent: Uses the @mariozechner/pi-coding-agent SDK and supports multiple AI providers. Set the API key for the desired provider as an environment variable, or you can store the key in ~/.pi/agent/auth.json (which takes precedence over environment variables). For the full list of supported providers and subscription services, see the Pi Agent documentation.

Codex: Install Codex CLI and set the OpenAI API key. Codex uses OpenAI models (for example o4-mini, gpt-4.1, codex-mini); see the Codex documentation for details.

Starting the Development Environment

In development mode, you need to install the corresponding dependencies for the framework you use: Claude Code (only when using the claude_code framework), OpenCode (only when using the opencode framework), Codex CLI (only when using the codex framework). By default, SQLite is used for storage, and agents run as local processes with no external services required. All data is stored in ~/.shire/.

Authentication Configuration

By default, Shire does not enable authentication, which is suitable for local development. To protect a shared or remote instance, set both SHIRE_USERNAME and SHIRE_PASSWORD. Once enabled, the dashboard displays a login page, and all API/WebSocket connections require a valid JWT token; the signing key is automatically generated on first startup at ~/.shire/.jwt-secret.

Build and Release

Shire is distributed via npm as standalone binaries with no runtime dependencies. The build command compiles the backend into a standalone binary and pre-builds the frontend assets. Releases are automated through GitHub Actions; tagging triggers the workflow, which builds and publishes the binaries for each platform: @agents-shire/cli-darwin-arm64 (macOS Apple Silicon), @agents-shire/cli-darwin-x64 (macOS Intel), @agents-shire/cli-linux-x64 (Linux x64), @agents-shire/cli-linux-arm64 (Linux ARM), @agents-shire/cli-win32-x64 (Windows x64), as well as the meta-package agents-shire that installs the correct binary.

Use Cases

Suitable for users who want AI agents to persist over the long term, accumulate work results across sessions, and allow multiple agents to collaborate autonomously; suitable for teams or individuals who need to organize agents by project, share files and context, and set up scheduled tasks and alert notifications; also suitable for developers who want the freedom to choose the underlying model provider and run agents locally without relying on external services.

Limitations

The material does not provide specific pricing, cost, performance metrics, or effectiveness data. Authentication is disabled by default, and you need to configure the username and password yourself when using it for a shared or remote instance. Different runtime frameworks have different configuration requirements for model formats and API keys, and you need to set them up according to the corresponding documentation.

Related tools

Browse category