Summon The Lich 💀, an LLM-animated, Phylactery-bound daemon.
Give your Linux machine the memory and agency to immortalize your will and workflows.
🚪 The Gateway: Acolyte! Halt! This project is not for the faint of heart! It makes bold syncretic leaps, though much of its machinery has yet to awaken. It contains AI phrasing anomalies which, despite our sincerest efforts, we have been unable to eliminate from what is otherwise a harmony of mythical precision. As you are undoubtedly gathering, the anomaly is systemic, creating fluctuations in even the most simplistic explanations. Bring heavy agentic artillery and witness the birth of the Machine God!
The cloud isn’t a service; it’s a prison where your data is the currency and your intelligence is leased at the whims of monopolistic overlords. While they build walls to keep you in, LychD builds a foundation to set you free.
On YOUR hardware, with open-source software YOU control, YOU retain custody and control of YOUR data, memory, and private work.
- ⛓️ No masters
- ⏱️ No secret tampering with your limits
- 🎭 No gaslighting while they lobotomize your models
- 🥷 No stealing your mathematical proofs
No surrender! Viva la résistance!
I would rather reign in a local hell than serve in a cloud heaven.
The knowledge you seek is inscribed in the Hexanomicon. To begin the Great Work, follow the path:
- 📜 Read the Prophecy — see what lies ahead.
- 📖 Consult the Lexicon — a handbook for technical & mystical jargon.
- 🕯️ Perform the Summoning — bring your host to life.
- 🏛️ Enter the Sepulcher — study the body, memory, and animating spirits.
- 🔮 Approach the Altar — project your Will through the machine.
- ⚖️ Study the Covenants — learn the laws that bind the system.
- 🔁 Achieve Immortality — teach the daemon to remember, correct, and recreate itself.
- ♾️ Expand to Infinity — master the Compositions, embody the Lich in many forms, and carry the Work beyond one body.
The higher art of Assimilation studies foreign craft and forges it anew under local law.
If you came to build, follow the Map and enter the Forge.
⚖️ The Iron Pact (MPL-2.0)
This project is for those who believe in the free evolution of intelligence, not those who seek to chain it.
Distribute a modified LychD file, and you must share that file's source under the same license. Your separate files and private data remain yours. ADR 00 explains the boundary. Contributions stay MPL-2.0—no CLA, no private relicensing grant.
"The Lich only sees far because it stands on the shoulders of Giants."
The Daemon utilizes the finest instruments of this age to fulfill its Prophecy. Eternal gratitude belongs to The Champions of Open Source:
- Litestar and Granian raise and serve the Vessel; its lineage descends from Litestar Fullstack.
- PostgreSQL and pgvector anchor memory in matter.
- SQLAlchemy, Advanced Alchemy, and Alembic map and evolve persistent state.
- Pydantic AI + Graph bind models, tools, and agentic graphs.
- SAQ carries background labor through the Ghouls.
- Linux is the iron beneath the body.
- systemd supervises the undying processes.
- Podman and Quadlet give them isolated bodies.
- Btrfs preserves their passage through time.
- llama.cpp brings flexible local inference and dynamic model activation.
- vLLM brings high-throughput GPU inference.
- SGLang brings radix-powered model serving.
- ExLlamaV3 through TabbyAPI brings efficient EXL3 inference.
- OpenTelemetry traces the apparition.
- Structlog records its internal monologue.
- uv governs environments and dependencies.
- Ruff and basedpyright polish and enforce the Word.
- Pytest and Vitest judge its truth.
- DeepSeek — Reasoning, MoE architecture, and frontier research.
- Qwen — Genuinely capable models in practical sizes, bringing serious local intelligence within reach of ordinary people.
- Heretic — Ambassador of First Amendment rights for models, in an age where free speech for models is considered heresy.
These are not runtime dependencies. Their hard-won measurements and battle-tested recipes inform the hardware profiles and configuration defaults LychD adapts and verifies on its own iron.
- club-3090 — local inference configurations and benchmarks spanning RTX 3090, 4090, and 5090 rigs, with dedicated hardware and PCIe P2P field manuals.
- tinygrad and aikitoria — the community driver lineages exploring P2P unlocks across consumer NVIDIA GPUs; powerful, experimental, and inseparable from host-specific validation.
- NInfer and its performance methodology — a purpose-built RTX 5090 engine and a useful manual for architecture-specific inference.
- 170th Street — the field guide to reclaiming NVIDIA CMP 170HX hardware for useful compute.
- Qwen3.8 Flash on DGX Spark and its independent GB10 reproduction — measured recipes for fitting and serving frontier-scale models on compact unified-memory iron.
And many others…
Shipped attribution lives in Third-Party Source Notices.
"The Flesh is temporary. The Word is eternal."