AGPL-3.0 License with Visible Attribution Requirement
Copyright (c) 2026 Ocean Bennett
Hematite — a local AI coding harness and workstation assistant engineered specifically
for single-GPU consumer hardware, tested and developed on an NVIDIA RTX 4070 (12 GB
VRAM) — including its agent harness architecture, hybrid RAG retrieval system (The
Vein), terminal TUI, tool registry, workflow mode engine, context compaction system,
recovery recipe layer, swarm orchestration, hardware-aware GPU runtime, PageRank-powered
repo map, L1 hot-files context layer, session memory and cross-tool import pipeline,
sandboxed code execution engine, system administration and workstation inspection layer,
self-contained voice/TTS pipeline, document and image attachment system, MCP server
integration, and all source code in this repository — is licensed under the GNU Affero
General Public License v3.0, with the additional visible attribution requirement below.
Hematite is purpose-built for the constraint that matters in practice: one local GPU,
one coding model, and one co-resident text embedding model — all running simultaneously
on a single consumer GPU with no cloud, no per-token billing, and no pretending that a
10 GB consumer card is a cloud worker fleet. The dual-model architecture (coding model
+ embedding model on the same GPU for hybrid semantic retrieval) is a core part of what
makes Hematite distinct. That architecture — local-first, hardware-honest, dual-model,
operator-visible — is the core of what this project is.
Hematite has grown beyond a pure coding assistant into a general-purpose local AI
workstation harness: it can inspect and diagnose the host machine (PATH, processes,
services, network, disk, toolchain health), execute sandboxed JavaScript/TypeScript
and Python in subprocesses, speak responses aloud via a baked-in voice engine with no
cloud dependency, and ingest cross-tool memory exports from other AI coding tools
(Claude Code, Codex, ChatGPT). These capabilities are part of the same AGPL-3.0
licensed work and subject to the same attribution and copyleft terms.
────────────────────────────────────────────────────────────────────────────────
ADDITIONAL TERM (permitted under AGPL-3.0 Section 7):
────────────────────────────────────────────────────────────────────────────────
Any tool, plugin, application, or service that uses, derives from, or is
substantially built upon this codebase or its architecture must include a visible
credit in one or more of the following locations:
- The tool's public-facing README or documentation
- The tool's in-app about panel or credits screen
- The tool's store page, landing page, or repository description
The credit must read, at minimum:
"Built on Hematite by Ocean Bennett
(https://github.com/undergroundrap/hematite-cli)"
or an equivalent that names Ocean Bennett and links to this repository.
This requirement applies regardless of how substantially the code was modified.
────────────────────────────────────────────────────────────────────────────────
Core license terms:
1. Attribution: Retain this copyright notice and the visible credit above in
any derivative work, fork, or project that uses Hematite or its architecture.
2. Copyleft: Derivative works must be released under AGPL-3.0. You may not
incorporate this code into a closed-source tool or product.
3. Network use: If you run a modified version as a hosted or networked service
(e.g., a cloud-based AI coding assistant built on Hematite's harness), you
must make the complete modified source code publicly available. Running a
hosted service counts as distribution.
4. Hardware intent: Hematite is designed around single-GPU consumer hardware
running a dual-model configuration (one coding model, one embedding model).
If you adapt it for multi-GPU, cloud-hosted, or enterprise-scale deployments,
the copyleft and attribution terms above still apply in full.
5. Scope: The AGPL-3.0 terms apply to the full scope of this software, including
but not limited to the AI agent harness, workstation inspection tools, sandboxed
execution engine, voice pipeline, retrieval system, session memory, and any
future capabilities added to this repository.
6. Commercial use by the original author: Ocean Bennett, as copyright holder,
reserves the right to release commercial versions of this software under
separate licensing terms, including custom builds for specific hardware
configurations or enterprise deployments. This does not affect the open-source
rights granted to the community under this license.
Full AGPL-3.0 license text: https://www.gnu.org/licenses/agpl-3.0.en.html