A private AI that runs on your hardware. No cloud required, no data leaves your network. From a personal assistant on your laptop to a full AI platform for your organization.
Install on your laptop or desktop. Takes 5 minutes.
Like ChatGPT, but running entirely on your computer. Ask questions, brainstorm ideas, get help writing emails or code — without sending a single word to the cloud. Your conversations stay yours.
Ollama, vLLM, llama.cpp, or 25+ other backends. Supports streaming responses, conversation history, session export. Choose from models like Llama 3, Mistral, Qwen, DeepSeek, and more.Point Eldric at a folder of PDFs, Word docs, or text files and ask questions in plain English. It reads, understands, and finds the answers — even across hundreds of documents.
SQLite, FAISS, or ChromaDB). Documents are chunked, embedded, and indexed locally. Semantic search with configurable chunking strategies and embedding models via Ollama or OpenAI.Get help with blog posts, reports, translations, cover letters, or social media content. Eldric understands context, matches your tone, and helps you overcome writer's block — all privately.
prompt templates for repeatable tasks, session persistence for long-form writing, and export to multiple formats.Speak to Eldric and hear it respond — like a voice assistant, but private. Great for hands-free work, accessibility, or when you just prefer talking over typing.
Whisper.cpp for speech-to-text, any LLM for reasoning, Piper TTS for natural speech output. All local, GPU-accelerated, streaming via SSE/WebSocket.Set up a shared AI for your office or department.
Set up one Eldric server and give everyone in your office their own private AI assistant — through a simple web browser. No cloud subscriptions, no per-seat fees. Your team's conversations and data never leave your building.
/chat with TLS, API key auth, and rate limiting. Multi-tenant isolation via Agent Worker. Supports OpenWebUI integration, SSO, and role-based access. Deploy on a single GPU server or scale across multiple machines.Feed Eldric your handbooks, SOPs, product docs, and meeting notes. Now anyone on the team can ask questions and get accurate answers from your company's own knowledge — instantly.
FAISS, ChromaDB, or built-in SQLite vectors. Knowledge bases are isolated per team or department with configurable access.Connect your email, WhatsApp, Teams, or SMS. Eldric reads incoming messages, classifies them, suggests replies, and can auto-respond to routine questions — with your approval.
IMAP/SMTP, WhatsApp Business API, Signal, Twilio SMS, Microsoft Teams, and XMPP. AI auto-response with approval queue. All messages stored locally with semantic search via Data Worker.Turn your home into a smart, AI-powered space.
Eldric connects to your smart home devices — lights, thermostats, cameras, sensors — and lets you control everything with natural language. No cloud service needed, so it works even when the internet is down.
MQTT, Zigbee, Z-Wave, ModBus, and OPC-UA. Supports device discovery, automation rules, and AI-powered scene management. Works with Home Assistant, Philips Hue, and hundreds of IoT devices. Dashboard at /dashboard.Enterprise-scale AI infrastructure you fully control.
Access NASA satellite data, CERN particle physics, genome databases, climate data, drug interactions, and more — all from one interface. Eldric connects to 140+ scientific data sources so researchers can focus on discovery, not data wrangling.
BLAST, NCBI, UniProt), pharma (AlphaFold, ADMET, molecular docking), CRISPR guide design, LIMS sample tracking, quantum computing simulation, and 15+ space agency APIs. GLP and 21 CFR Part 11 compliance modes.Define a goal and let a team of AI agents figure out how to accomplish it. One agent researches, another writes code, a third reviews it, and a fourth deploys. They coordinate automatically.
Start with a single machine and grow to hundreds of GPU nodes across data centers. Eldric handles load balancing, failover, multi-tenant isolation, and AI-powered routing — so your teams get fast, reliable AI no matter the load.
Edge (TLS/auth), Router (AI-powered load balancing), Workers (inference), Controller (orchestration). Supports 25+ backends including Ollama, vLLM, TGI, NVIDIA Triton, OpenAI, Anthropic. Enterprise licensing with hardware binding and multi-region deployment.Fine-tune AI models on your company's data — legal documents, medical records, engineering specs, customer interactions. The result is an AI that speaks your language and knows your domain better than any generic model.
Unsloth, Axolotl, TRL, DeepSpeed, MLX, and llama.cpp. Methods include LoRA, QLoRA, SFT, DPO, RLHF, PPO, and full fine-tuning. Training chains with visual node-based configuration. Multi-GPU and distributed training support.Automatically transcribe meetings, extract action items, detect speakers, and make all your audio and video content searchable. Process thousands of hours without sending anything to the cloud.
Whisper.cpp STT, Piper TTS, FFmpeg video processing. Speaker diarization, scene detection, frame extraction. All media indexed in Data Worker for semantic search across recordings. Streaming via SSE/WebSocket.Build AI-powered apps with a platform that gets out of your way.
Eldric exposes an OpenAI-compatible API. If your app already talks to OpenAI, just change the URL and you're running locally. No code changes. Works with LangChain, LlamaIndex, AutoGen, and every OpenAI SDK.
/v1/chat/completions and /v1/models API. Streaming via SSE. Native tool calling pass-through. Supports Ollama, vLLM, TGI, llama.cpp, MLX, NVIDIA Triton, OpenAI, Anthropic, Groq, xAI, Together, and more. Switch backends without changing client code.Ingest documents, generate embeddings, store vectors, and query with semantic search — all through REST APIs. Build knowledge-base chatbots, document Q&A, or context-aware assistants without managing separate infrastructure.
SQLite, FAISS, ChromaDB). Agent Worker adds agentic RAG with ReAct pattern, query decomposition, and multi-agent execution. REST APIs for /api/v1/vector/* and /api/v1/agent/*. Per-tenant namespace isolation.Use the DevToolbox to step through agent tool calls, test prompts interactively, profile token usage and latency, and generate test cases automatically. Stop guessing why your AI does what it does.
Eldric speaks the Model Context Protocol. Connect to any MCP server — databases, file systems, APIs, custom tools — and your AI can use them natively. Build once, use with any MCP-compatible client.
~/.config/eldric/mcp_servers.json. Supports stdio and SSE transports. Tools from MCP servers appear alongside native tools in agent workflows.Design, build, test, and deploy AI agents — without writing infrastructure code.
Describe what you need in plain English and the Agent Builder designs the architecture, generates the code, selects the right tools, and creates test cases. It's an AI that builds AI agents for you.
/agent-builder deploy.Need 50 agents for different departments? The Agent Generator uses domain templates to produce consistent, production-ready agents instantly. Pick a domain, set constraints, generate. Deterministic, repeatable, fast.
/api/v1/agent-generator/*.Combine 13 built-in agent types into coordinated swarms. An Explorer finds files, a Coder writes the fix, a Runner executes tests, and a Planner tracks the overall goal. Define the mission, agents handle the rest.
port 8885) coordinates goals, Agent Worker (port 8893) handles execution. UAP (Universal Agent Protocol) for inter-agent communication.Extend Eldric with server-side Python plugins (tools, filters, pipes) and client-side JavaScript widgets. Add custom tools the LLM can call, pre/post-process messages, or create entirely new virtual model backends.
Tool (server-side, callable by LLM), Filter (inlet/outlet message processing), Pipe (virtual model backends), Action/Widget (client-side JS). Plugins run in Python subprocesses via JSON-RPC. Manifest + valves configuration.Ship AI products on infrastructure you own. No cloud bills, no vendor lock-in.
Use Eldric as the backbone for your AI product. Multi-tenant isolation, per-customer API keys, usage metering, rate limiting, and a white-label web UI — everything you need to sell AI services without building infrastructure from scratch.
/v1/chat/completions API with streaming.Offer Llama, Mistral, Qwen, DeepSeek, or any open model — plus route to OpenAI or Anthropic when customers need it. AI-powered routing picks the best model and worker for each request automatically.
round_robin, least_connections, load_based, latency_based, and AI autonomous routing modes. AI routing uses an LLM to select workers based on model capabilities, load, and request characteristics. Workers support Ollama, vLLM, TGI, llama.cpp, MLX, Triton, and all major cloud APIs.Use the Agent Builder and Generator to create domain-specific AI agents, package them with knowledge bases and safety guardrails, and deploy them for customers. Version, A/B test, and iterate — all on your infrastructure.
/agent-builder deploy. Per-tenant isolation, knowledge base binding, safety constraint templates.Let your customers fine-tune models on their own data — through your platform. Training pipelines with visual configuration, automatic dataset preparation, and deployment to inference workers when done.
Unsloth, Axolotl, TRL, DeepSpeed, MLX, llama.cpp. Training chains with visual node-based config. Latent reasoning techniques (COCONUT, Quiet-STaR, Pause Tokens). Auto dataset preparation from knowledge bases via Agent Worker. Multi-GPU and distributed training.Add AI services to your portfolio. Eldric scales from one rack to entire data centers.
Deploy Eldric across your GPU fleet. Workers auto-register with the controller, routers balance load across them, and the Edge provides a single HTTPS endpoint for all your customers. Add or remove nodes without downtime.
Controller (port 8880) manages cluster, Router (port 8881) load balances, Workers (port 8890) run inference. SSH-based deployment via Controller API. Enterprise licensing with hardware binding, per-backend worker limits, and multi-region support. Edge farm mode for horizontal gateway scaling.Eldric's built-in licensing system lets you gate features by tier, enforce node limits, bind to hardware, and track usage. Integrate with your billing system or use the license server directly.
octopus.core.at with REST API. Feature gating across all workers (science, training, media, comm, agent).Purpose-built for scientific computing, research, and industrial applications. 140+ science APIs, distributed inference, xLSTM models.
Eldric is free for personal use. Download it now and have your own AI running in 5 minutes.