EULLM Engine: EU-Hosted Ollama-Compatible Local Inference
Tier C BYO local upstream — Ollama/OpenAI peer with audit JSONL; Forge + Hub still early.
Tier C watchlist BYO upstream — local GGUF Engine with Ollama + OpenAI APIs and audit JSONL; not a Galaxy COD or campus stack path.
Key Facts
- Tier C watchlist — Engine usable today; Forge/Hub early (vendor status).
- Apache 2.0; permissive-base models (Qwen/Mistral); Meta Llama excluded per policy.
- Drop-in Ollama port 11434 + OpenAI /v1 on the same binary.
- Copperway BYO upstream path only — not Galaxy COD, not campus CapEx stack.
- Do not underwrite tok/s or Z2 pricing against EULLM consumer/ARM demos.
What EULLM does
EULLM (github.com/eullm/eullm) is an Italy-based Apache 2.0 project shipping EULLM Engine: a Rust binary wrapping llama.cpp that runs any GGUF model, serves Ollama-compatible routes on :11434 plus OpenAI /v1/chat/completions, optional embedded chat UI on :11435, and a default local JSONL audit trail under ~/.eullm/audit/. Documented roadmap components include Forge (Python prune→distill→quantize→LoRA→GGUF pipeline) and Hub (EU model registry) — both described as early / not production-ready in vendor docs. Infrastructure claims point at Hetzner DE / OVH FR hosting with zero telemetry outside the EU (vendor claim). Meta Llama weights are excluded per their licensing policy; permissive-base models (Qwen, Mistral, etc.) are the stated focus.
What they stand for
EU-hosted local inference with operator-controlled transparency — Apache 2.0 runtime, optional local audit trails, and no vendor telemetry outside European infrastructure (per vendor claims). A BYO alternative to US-hosted inference APIs for teams that need localhost or on-prem OpenAI-compatible serving.
Supply-chain fit for AGICY
Possible Copperway BYO upstream or staff demo runtime via OpenAI SDK (base_url=http://localhost:11434/v1) — explicitly not a Galaxy / Tenstorrent COD path. AGICY documents third-party options buyers may self-host on laptops, edge, or non-campus hardware; Copperway remains the commercial gateway product and the Vasilikos campus remains pre-construction. Tier C honesty: do not place EULLM in the Phase 1 COD serving shortlist alongside Moreh-class vendors. Forge/Hub stay off campus stack planning until dated production distribution. Cross-read the selective EU open-source catalog for adjacent tooling.
Status
Not a partner today · watchlist only · Tier C BYO upstream · audit JSONL is operator tooling, not conformity certification
Latest news angles (2025–2026)
| When | Signal | Source |
|---|---|---|
| 2026 | EULLM — repository and Engine releases | GitHub |
| 2026 | EULLM architecture — Engine, Forge, Hub | EULLM docs |
FAQ
How should AGICY consider EULLM in outreach?
As a documented Tier C BYO local upstream option for Copperway-adjacent buyers — not as CapEx, campus serving, tenancy, or Galaxy COD. Useful for EU AI Act transparency narratives on localhost; not a Phase 1 stack primary.
Does AGICY ship EULLM?
No. EULLM is a third-party Apache 2.0 project. AGICY may document it as a self-hosted OpenAI-compatible upstream. Copperway remains the gateway product; the campus remains pre-construction.
Is a local audit trail EU AI Act compliance?
No. Append-only request/response logs support operator transparency work. They are not an EU AI Act conformity assessment, notified-body assessment, or high-risk system certification.
Can we use EULLM numbers for Galaxy pricing?
No. Consumer/ARM demos and GGUF benches are not AGICY measured COD tok/s or Z2 Unit Economics.
Related intelligence
Official & primary sources
- github.com/eullm/eullm
- Getting started
- Architecture — Engine, Forge, Hub
- GitHub: EULLM — repository and Engine releases
- EULLM docs: EULLM architecture — Engine, Forge, Hub
- AGICY stack ecosystem shortlist (internal research archive).
Explore the Stack Ecosystem
Illustrative mapping of serving, tenancy, assurance, and orchestration vendors that could complement AGICY's Cyprus Galaxy campus — potential affiliates, not current partners.
View Stack Ecosystem