open-index: Build Agent Memory on SQLite and OpenSearch

Overview

Open Index is a deterministic memory layer for domain-specialized agents. It organizes a domain as a searchable context graph: you define concepts, fields, entities, and their relationships, then agents can retrieve and update that structured knowledge instead of relying only on unstructured context. SQLite with FTS5 is the default storage and search backend; OpenSearch is available for shared, multi-writer deployments.

The project is aimed at teams building agents for areas such as support, sales, infrastructure, or other domains where knowledge needs to stay structured and current. Its CLI, explorer UI, and MCP server provide ways to build, inspect, and connect a brain to an agent. See the repository README and project site for details.

Key Features

• Define domain concepts with YAML schemas, including field types, search behavior, and per-field ranking boosts.
• Represent entities and their connections with typed records and related_to edges.
• Keep curated entities in Git-tracked files, while storing connector-generated or agent-written entities in the index.
• Search through the CLI or MCP, with MCP tools for reading and writing validated entities and schemas.
• Ingest data from connectors and bulk-import JSON, JSONL, or CSV files.
• Explore relationships, search activity, and connector jobs through the UI.
• Choose SQLite with FTS5 for local or single-writer use, or OpenSearch for shared deployments and larger-scale search features.

Use Cases

• A support team can model products, issues, and customer segments so an agent can retrieve connected troubleshooting context and record new knowledge.
• An infrastructure team can connect services to runbooks, dashboards, and alerts for operational agents.
• A sales team can maintain structured lead, deal, and account information for agents that need domain-specific context.
• A team with multiple agents writing to shared knowledge can serve a brain over HTTP and use OpenSearch as the backend.

Getting Started

Install the CLI and optional UI and MCP components, then try a bundled example:

pip install -e '.[all]'
open-index index --brain examples/support-brain
open-index ui --brain examples/support-brain

For setup options and deployment details, see the README.

Alternatives

• a-mem-mcp: A-MEM focuses on evolving, connected agent memories, while Open Index provides a domain-agnostic context graph with SQLite or OpenSearch backends.
• Memary: Memary emphasizes personalized memory for autonomous agents, while Open Index centers on structured concepts and relationships managed through MCP.

Considerations

• The project is less than six months old and is changing quickly. It has recent commit and pull-request activity, but no releases in the past year, so teams should expect some adoption risk.
• Development is not concentrated in one contributor, though the project has a small contributor base. The health data records no issues closed in the recent period, so issue responsiveness is not established by that measure.
• Python 3.10 or newer is required. Docker and Compose are available for deployment but are not prerequisites for local use.
• SQLite is single-writer. The README recommends OpenSearch when multiple agents need to write to a shared brain; that adds a service to operate.
• Semantic search requires the optional semantic installation extra. The README notes that SQLite uses brute-force cosine scans and recommends OpenSearch beyond roughly 10,000 entities.
• The MIT license permits broad use, subject to its license terms. The README also notes that type-level and temporal search boosters are not implemented.