Agience Documentation

Status: Reference Date: 2026-04-10

Agience is a human-in-the-loop knowledge curation platform — an OS-like substrate where AI agents and humans collaborate over structured, versioned information through open MCP standards.

---

New to Agience?

• What is Agience? — Start here: core concepts, goals, and the information triangle
• Quickstart — Get running in minutes

---

Overview

Document	Description
What is Agience?	Core concepts, goals, and the information triangle
Platform Overview	High-level architecture, layer model, and key components
Information OS Analogy	How Agience relates to an operating system for knowledge
Manifesto	The principles and philosophy behind Agience
Solution Taxonomy	How solutions, personas, and content types are organized

---

Use Cases

Document	Description
Use Cases	Overview of all documented use cases
Email Management	Intelligent email triage and response with AI agents
Personal Assistant	Telegram-based personal AI assistant
Meeting Intelligence	Capturing and surfacing knowledge from meetings
Executive AI Team	Multi-agent coordination for executive workflows
Event Planning	AI-assisted event coordination and logistics

---

Getting Started

Document	Description
Quickstart	Fastest path to a running Agience instance
Local Development	Full dev setup with Docker, backend, and frontend
Self-Hosting Guide	Run Agience on your own infrastructure
Deploy to EC2	Step-by-step EC2 deployment walkthrough
Admin Setup	First-run admin configuration and provisioning

For MCP client configuration, see MCP Client Setup.

---

Architecture

Document	Description
Architecture Overview	System architecture: services, databases, and communication patterns
Layered Architecture	Core / Handlers / Presentation layer boundaries and rules
Artifact Model	Artifact structure, versioning, references, and ArangoDB lifecycle
Content Types	MIME-based content type system and handler ownership model
Security Model	Auth, JWT, API keys, scopes, grants, and trust model

---

Features

Document	Description
Anchored Reasoning	Premium tier — anchors AI answers to verified evidence, prevents hallucination (patent-pending)
Search	Hybrid BM25 + semantic search, aperture control, query language
Agent Execution	Agent artifacts, operator execution, and invocation API
Desktop Host Relay	Local MCP bridge via WebSocket relay for desktop apps
Workspace Automation	Event handlers, workspace triggers, and automation patterns
External Auth	Connecting external identity providers (OAuth, OIDC)
Content Type Registry	How content types are registered, resolved, and served

---

MCP Integration

Document	Description
MCP Overview	How Agience acts as both MCP server and MCP client
Client Setup	Connecting an MCP client to Agience
Client Instructions	Tool surface reference for external MCP clients
Server Development	Build and register MCP servers: auth, callbacks, content types
Testing	How to test MCP tools and server connections
VS Code	Using Agience as an MCP server inside VS Code

---

Reference

Document	Description
Vocabulary	Canonical term definitions (Artifact, Card, Operator, etc.)
API Reference	Backend REST API endpoint reference

---

Guides

Document	Description
Best Practices	Coding conventions, testing, and anti-patterns
Component Guide	Frontend component inventory and usage patterns
Color Standards	Design system color palette and usage rules
Run, Release & Deploy	Operator runbook: run locally, cut releases, deploy hosted sites

---

Contributing

Document	Description
Contributing	How to contribute code, docs, and content types
Changelog	Release history and version notes