halofyai/halofy

What it solves

Halofy provides a centralized governance and access layer for AI agents within an organization. It prevents agents from needing their own separate identity systems, policy engines, or audit logs, ensuring that every agent has the correct organizational context and follows a unified security policy.

How it works

Halofy acts as a middleware between organizational knowledge and AI agents. It uses a server-owned identity model where credentials resolve the actor, role, and namespace server-side, ensuring that agents cannot assert their own identity. Access is managed through a namespace-based ACL (Access Control List) and role-based decisions that apply across all interfaces (MCP, HTTP).

For storage, it uses Postgres/pgvector as the primary authority for context, embeddings, and audit logs. It supports a tiered storage approach with encrypted, Git-versioned knowledge for cold storage. Retrieval is handled by read-only drivers that operate within a scoped view, ensuring they cannot write to the storage or bypass security policies.

Who it’s for

Organizations deploying multiple AI agents and applications that require strict access control, auditable provenance of knowledge, and verifiable erasure of data (signed erasure).

Highlights

• Unified Governance: Single identity and policy model for all agents, applications, and operators.
• MCP and HTTP Support: Provides a governed access surface via Model Context Protocol (MCP) and HTTP APIs.
• Organization-wide ACLs: Server-resolved identity and namespace-based access control to prevent unauthorized data access.
• Auditable Lifecycle: Append-only history with provenance, entity resolution, and signed erasure certificates.
• Open Infrastructure: Self-hostable, offline-ready, and uses Postgres as the authority with no telemetry.
• Connectors: Includes public connectors for filesystem, Postgres, Obsidian, and CSV imports.