agience-core

Overview

Agience operates as an 'operating system for AI workflows', transforming AI-generated outputs into durable, governed knowledge artifacts with identity, provenance, and version history. The platform provides both MCP server capabilities (exposing tools to various clients) and MCP client functionality (consuming other vendor servers). Its architecture includes typed artifact objects, versioned collections, and human-in-the-loop approval gates as first-class operators. The system uses ArangoDB for storage and implements multi-provider OAuth2 authentication with scoped API keys.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• create_artifact

  Create a new typed artifact with version history

  Try:Create a decision artifact summarizing the project risksTry:Generate a new action item from the meeting transcript
• query_artifacts

  Search and retrieve artifacts using query language

  Try:Find all decision artifacts from last weekTry:Search for artifacts tagged as 'urgent'
• workspace_commit

  Commit workspace artifacts to a collection

  Try:Publish the current workspace contents to the team collectionTry:Commit draft artifacts for review
• retrieve_artifact

  Get a specific artifact by its ID

  Try:Show me artifact ID 1234Try:Retrieve the decision artifact from yesterday
• update_artifact

  Modify an existing artifact with version tracking

  Try:Update the action item with new statusTry:Add a comment to the decision artifact
• create_collection

  Create a new collection for organizing artifacts

  Try:Create a new project collectionTry:Set up a team workspace for Q4 planning
• search_artifacts

  Perform semantic search across artifacts

  Try:Search for artifacts about customer feedbackTry:Find similar project risk artifacts
• list_workspaces

  List all available workspaces

  Try:Show me all my workspacesTry:List team workspaces
• ingest_transcript

  Process a meeting transcript into structured artifacts

  Try:Ingest yesterday's team meeting transcriptTry:Process the earnings call recording
• export_collection

  Export a collection for external use

  Try:Export the project collection to PDFTry:Download the Q3 decisions as JSON
• grant_permissions

  Set access permissions for artifacts and collections

  Try:Grant read access to the team collectionTry:Set up API key for external service
• get_provenance

  Retrieve provenance chain for an artifact

  Try:Show me the history of artifact 1234Try:Trace who created this decision

Note: Tools inferred from MCP documentation mentioning '11 tools at /mcp' and architecture descriptions of artifact handling, along with common MCP tool patterns for artifact management systems.

Comparable tools

Installation

# Linux/macOS
curl -fsSL https://get.agience.ai/home/install.sh | sh

# Windows (PowerShell)
irm https://get.agience.ai/home/install.ps1 | iex

git clone https://github.com/Agience/agience-core.git
cd agience-core
# Linux/macOS
./agience dev -f --build
# Windows
agience dev -f --build

"mcpServers": {
  "agience": {
    "command": "python",
    "args": ["-m", "agience.mcp"]
  }
}

FAQ

What types of MCP tools does Agience provide?

Agience provides 11 MCP tools focused on artifact management, search, ingestion, reasoning, and output generation, accessible via `/mcp` endpoint.

How does Agience handle AI output provenance?

Agience structures provenance as infrastructure - committed artifacts carry records of what produced them, from what inputs, under whose authority, with this being a fundamental architectural feature rather than an add-on.

Compare agience-core with