Launch HN: Hyper (YC P26) – An enterprise brain for agent development

To solve the performance bottleneck caused by AI agents failing to understand fragmented internal company information (Slack, documents, email, etc.), Hyper has launched, integrating company data and managing it in the form of a Knowledge Graph.

Core Technology and How It Works

• Hybrid memory system: Composed of Episodes, which preserve the original data, and Facts, which extract meaning from the data. Facts are stored in a 'subject-predicate-object' structure, and relationships between pieces of information (contradiction, derivation, replacement, etc.) are defined to maintain context.
• Staying current: When new information comes in, it updates the graph by replacing existing information, while preserving previous records so that the history of decisions can be tracked.
• Sophisticated search and permission management: Combines Query-expansion, Semantic search, and Postgres full-text search using Reciprocal Rank Fusion (RRF) to improve search accuracy. It also reflects each user's access permissions, generating answers only within the information each user is allowed to see.

Integration into Agent Workflows

• Through the lifecycle hooks of major AI coding tools such as Claude Code, Cursor, and Codex, relevant context is automatically injected into prompts, and new facts are extracted from responses.
• Supports MCP (Model Context Protocol), enabling information access and expansion through tool calls.