# Halofy

> An open-source AI ops control plane that gives every agent and team member scoped, governed access to organizational knowledge through one identity and policy model.

Halofy is an open-source access and governance layer for AI agents across an organization, built by Hivenix Inc. and based in San Francisco. The core kernel is licensed under AGPL-3.0 and available on GitHub, while a managed cloud service (currently in early access alpha) layers hosted memory, LLM orchestration, and channel integrations on top. The project sits between organizational knowledge and the agents that consume it, resolving identity, enforcing access, governing change, recording every outcome, and making erasure independently verifiable.

## What It Is

Halofy is an AI ops control plane — a single governance boundary through which every AI agent, human operator, and automated workflow in an organization accesses shared context. Rather than each team wiring its own models, credentials, data sources, and channels independently (what the product page calls "ungoverned AI sprawl"), Halofy centralizes identity, policy, context, usage attribution, and audit under one surface. It is model-agnostic and does not compete with model providers; it sits above Claude, GPT, Gemini, Grok, and others, routing tasks to whichever model a policy allows.

## How the Governance Architecture Works

The kernel enforces five hard boundaries documented in the repository:

- **Server-owned identity:** Namespace, actor, role, and source scope come from the credential, never from a request body. MCP tool schemas do not accept caller-asserted identity fields.
- **Organization-wide ACL:** The same access-control and role decisions apply across agents, applications, operators, MCP, and HTTP. A caller in namespace `org/support/agent-1` can see its own namespace and ancestors, but siblings and descendants remain hidden.
- **Governed context lifecycle:** Every write follows an auditable path with provenance, supersedence chains, and append-only history. Corrections close the prior fact's validity interval rather than overwriting it.
- **Delegated reads:** A retriever receives only an ACL-scoped `ScopedView` and can return references only — it cannot write or reach audit or storage modules.
- **Postgres as authority:** Context, embeddings, policy, provenance, tombstones, and audit live together. Embedded PGlite is the zero-setup local default for development and tests.

The open-source build passes 2,678 tests across 205 files and all 6/6 driver-conformance checks, according to the repository README.

## The Managed Service Layer

On top of the open kernel, Halofy's hosted service adds capabilities the repository explicitly marks as commercial:

- **Scoped badges** that travel with agents across Claude Code, Codex, Hermes, and other clients, carrying only the context a task requires.
- **LLM orchestration harness** — one endpoint routing across providers with policy-preserving model selection and spend attribution.
- **Halofy Assistant** deployable org-wide or as a custom harness, with multimodel orchestration.
- **Channel integrations** to Slack, Google Chat, Microsoft Teams, and Telegram, so the same assistant and guardrails reach teams where they already work.
- **Token spend attribution** by team, agent, and model, with real-time optimization suggestions.

The homepage states the platform has integrated 60+ business backends, recorded 45K+ memory writes, supports 15+ channels, and targets under one hour of setup time — these are vendor-published figures.

## Security Posture

The managed service is described on the product page as enterprise-grade by default:

- End-to-end PII detection, masking, and redaction before data reaches an agent.
- Complete data isolation — every business runs in its own isolated environment with no crossover.
- Zero data retention — Halofy states it never stores or trains on customer or business data.
- AES-256 encryption at rest and TLS in transit on every request.
- SSO and append-only audit available on the Team tier.

The open-source kernel includes a documented threat model covering known partial controls and adversarial cases.

## Open-Source Deployment Model

The standalone build is designed to run without any licence key, activation, or entitlement check. It ships with embedded PGlite as the zero-setup local default, optional Postgres/pgvector for durable deployments, and Docker Compose for self-hosting. No telemetry or usage beacon is included. Four public connectors ship in the repository: filesystem, Postgres, Obsidian, and manual CSV import. The driver interface is Apache-2.0 licensed so connectors can be embedded without relicensing the host application. The kernel and console are AGPL-3.0-or-later.

## Current Status: Alpha Launch

The homepage banner reads "Halofy Alpha is now Live" and the pricing page offers early access at no cost for current users. The GitHub repository was created in August 2026 and had 358 stars and 4 forks as of late August 2026. The open-source platform tier is listed as "Opensource Soon" on the pricing page, indicating the self-serve open-source path is still being finalized for general availability even as the managed alpha is live.

## Features
- Scoped identity and role-based access control for agents and humans
- Organization-wide knowledge base with provenance and append-only audit
- MCP and HTTP endpoints for agent connections
- LLM orchestration harness routing across Claude, GPT, Gemini, Grok, and more
- Halofy Assistant deployable org-wide or as a custom harness
- Channel integrations: Slack, Google Chat, Microsoft Teams, Telegram
- Token spend attribution by team, agent, and model
- End-to-end PII detection, masking, and redaction
- AES-256 encryption at rest and TLS in transit
- Zero data retention policy
- Complete data isolation per business
- Signed erasure with independently verifiable certificates
- Embedded PGlite for zero-setup local development
- Postgres/pgvector for durable self-hosted deployments
- Docker Compose self-hosting support
- Append-only audit log
- SSO support
- Backend connectors: Notion, Linear, GitHub, Google Drive, HubSpot, Obsidian, Jira
- Open-source plugin ecosystem
- Offline-ready kernel with no telemetry

## Integrations
Notion, Linear, GitHub, Google Drive, HubSpot, Jira, Obsidian, Slack, Google Chat, Microsoft Teams, Telegram, Claude Code, Codex, Cursor, Hermes, Grok, Mastra, Anthropic Claude, OpenAI GPT, Google Gemini, Azure OpenAI, Postgres, pgvector

## Platforms
LINUX, WEB, API, CLI

## Pricing
Open Source, Free tier available

## Links
- Website: https://halofy.ai
- Documentation: https://halofy.ai/docs
- Repository: https://github.com/halofyai/halofy
- EveryDev.ai: https://www.everydev.ai/tools/halofy