AGENTS.db is an advanced vectorized flatfile database designed specifically for LLMs to efficiently query and store context. Combining the advantages of a canonical knowledge base with the flexibility of append-only layers, it ensures clear provenance and supports fast local searches, all while retaining the integrity of your project's knowledge.
AGENTS.db is an advanced toolkit designed to enhance context management for large language models (LLMs). As a vectorized, flatfile database, it streamlines the storage and retrieval of context, making it an essential component for agent systems that require high performance and reliability.
Key Features
- Read-only, Canonical Knowledge Base: Serve as the foundational layer, ensuring that the core documentation remains immutable and dependable.
- Append-only Layers: Include user-generated notes, derived summaries, and proposals without altering the base layer.
- Provenance Tracking: Maintain clear records of authorship and source attribution for each data chunk, enhancing transparency.
- Optimized Local Search: Facilitate fast and efficient searches across multiple layers, ensuring that users can quickly access relevant information.
AGENTS.db builds upon the existing AGENTS.md standard by consolidating the functionality into a single, structured database format. It combats the fragmentation often faced with multiple files, allowing users to manage their knowledge effectively.
Layer Structure
AGENTS.db organizes knowledge into distinct layers:
- Base Layer: The
AGENTS.db, serving as the immutable source of truth. - User Layer: The
AGENTS.user.dbfor durable human additions, retaining important context. - Delta Layer: The
AGENTS.delta.dbfor proposed additions that can be reviewed before integration. - Local Layer: The
AGENTS.local.dbfor ephemeral notes that do not require source control commitment.
The layered system allows for a prioritized search where higher-precedence layers take precedence:
local > user > delta > base
Usage Scenarios
AGENTS.db is ideal for those who need to manage documentation and context collaboratively and transparently in dynamic environments. By utilizing append-only structures, users can easily track changes and revisions while ensuring that the foundational data remains intact.
Quick Commands
- **Initialize a New Database: **
agentsdb init
- Search Across Layers:
agentsdb search --base AGENTS.db --query "example query" -k 5
- Export Layers:
agentsdb export --dir . --format json --layers base,user,delta,local --out agentsdb-export.json
Web Interface & MCP Server
AGENTS.db also includes a web UI for easy interaction and management of layers as well as an MCP server that can be integrated with various programming environments, making it versatile for a range of applications.
For comprehensive guidance on workflows, specifications, and implementation, refer to the provided documentation located in docs/ directory within the repository.
No comments yet.
Sign in to be the first to comment.