OctaMem launches a memory layer for AI agents that stores and retrieves context without requiring a vector database, addressing the cost of repeatedly re-reading the same information and organizational knowledge loss.

What happened

OctaMem introduced a memory infrastructure that organizes agent data into three types—semantic (facts and knowledge), episodic (events and history), and procedural (workflows and rules)—and retrieves only relevant context per request. The system ingests documents (contracts, PDFs, spreadsheets, emails) and parses them into structured, queryable records rather than embedding blobs. It integrates via REST, Python, JavaScript, SDKs, IDE plugins, and MCP (a protocol for tool-calling) and is available on a free tier with 2 GB memory, no credit card required.

What to watch

OctaMem offers configurable retention windows, per-record deletion, role-based access control (Okta, Entra, Google SSO), AES-256-GCM encryption at rest, TLS 1.3 in transit, and multi-AZ deployment with 30-minute RTO and 5-minute RPO. A desktop app for macOS (Apple Silicon and Intel) is available now; Windows and Linux versions are coming. Every memory record carries an immutable ID, type, score, source, and timestamp, enabling full auditability and deletion by record or context.