Document-grounded chat, an agentic coding IDE, and a design studio — plus translate and MCP hubs — one interface, running entirely inside your own network. Your models. Your hardware. Your rules.
Air-gap capable — works with zero outbound connectivity
Seven product hubs share one login, one permission model, and one set of files. Chat, code, and design are the daily drivers — and they hand work off to each other.
Drop in a PDF, contract, or spreadsheet and ask questions against it. Retrieval runs on pgvector inside your database — the source text never leaves the machine it landed on.
Three modes, deliberately separated. Plan investigates and writes a spec without touching a file. Build implements it. Ask answers questions read-only. Every session gets an isolated workspace with its own git history.
Scope a design through guided questions, and Eleftheria writes a DESIGN.md contract plus a self-contained HTML prototype. Edit it on a multi-page canvas, then hand the whole thing to Eleftheria Code to build for real.
Run document and text translation on the same models you already host. Set languages, tone, and multi-pass workflows — without sending content to a third-party SaaS.
Connect Model Context Protocol tool servers so agents can call your internal APIs and databases under group permissions.
Organize chats, files, and code workspaces under shared project boundaries for teams.
Structured outputs from any hub — tables, charts, and documents you can export or reopen later.
Every component ships in your Docker Compose stack. Inference hits model endpoints you operate. Turn on air-gap mode and outbound access is disabled at the application layer.
Web search is disabled at the source, model weights load from local cache, and the data-analysis sandbox runs with networking switched off entirely.
Point Eleftheria at any OpenAI-compatible endpoint. Chat, embeddings, and reranking are configured independently, so you can mix model sizes per workload.
Each user gets their own coding workspace on disk. Spreadsheet analysis executes in a throwaway container with no network route out.
Bring your own OpenAI-compatible inference, govern what each group can call, and automate the platform with bearer tokens — local-first, with optional cloud providers when egress is allowed.
Point chat, embeddings, and reranking at vLLM, llama-server, Ollama, or any OpenAI-compatible endpoint you operate.
Local is the default. Optionally register OpenAI, Gemini, DeepSeek, Groq, or Azure when your policy allows outbound traffic.
Map LDAP groups to model catalogs and hub access so teams only see the inference you approve.
Issue wbr_ bearer tokens for REST and WebSocket scripting — the same auth boundary as the web UI.
Eleftheria is a multi-tenant deployment, not a desktop toy. Identity, permissions, and auditing are part of the product rather than an enterprise upsell.
Run a response past several verifier models and a judge, then surface a confidence rating next to the answer.
Curated document collections, maintained centrally and attachable to any conversation.
Authenticate against directory infrastructure you already run, and map existing groups straight onto Eleftheria roles.
Decide which groups reach which models, and gate the Code and Design workspaces independently per user.
Spreadsheets are handled by a pandas agent inside a container with no network access.
Conversations to Word, artifacts to PDF and Excel, designs to PNG, PDF, and HTML, projects to ZIP. Nothing is trapped in the product.
Register Model Context Protocol servers once and expose them to chat and coding agents under the same permission model.
An honest comparison. Hosted suites ship faster and have bigger models. Eleftheria gives you the one thing they structurally cannot.
On-premise licence for European teams. You run the hardware and models — we licence the workspace. Prices in EUR, net of VAT.
Core platform for small teams
Add AI translation workflows
All four modules, one platform
Unlimited scale, dedicated support
~€20 / user / month effective across paid tiers. Module add-ons from €3–€6 / user / month. LLM inference costs are yours — local or cloud providers you configure. Annual billing default; monthly +15%.
One deployment. Seven hubs. Nothing leaving the building.
