viki-neural-forge

Local LLM · Ollama · VIKI Neural Forge · self-hosted AI · private assistant

viki-neural-forge is an Ollama model image produced by VIKI (Virtual Intelligence Knowledge Interface) Neural Forge: your base checkpoint (e.g. Qwen, Gemma, Llama-class, Mistral, DeepSeek, Phi) plus a baked Modelfile SYSTEM block—persona, reinforced lessons, and generation parameters. It is not a separate foundation model; reasoning traces, tool calling, vision, audio, and context length follow the base you set in FROM.

Use it for offline LLM chat, RAG-style behavior hints in the prompt, and a consistent VIKI voice when paired with the full Python agent (optional). Documentation: Orythix/VIKI-public.

Pull and run

ollama pull orythix/viki-neural-forge
ollama run orythix/viki-neural-forge

ollama show orythix/viki-neural-forge --modelfile

VIKI agent platform (what the project offers)

Running the full VIKI application (not this file alone) unlocks an autonomous AI agent, sovereign AI / privacy-first stack: local-first, air-gapped mode, Orythix cognitive architecture, judgment engine (reflex / shallow / deep), capability gating, ethical governor, reflection layer, SQLite lessons and semantic memory, failure memory, continuous learning, Neural Forge (viki-neural-forge / viki-evolved profile), multi-model routing, ensemble deliberation (optional), low-resource mode, prewarm for Ollama.

Interfaces: CLI, Flask REST API, SSE chat streaming, WebSocket missions, React + Vite dashboard, hologram voice UI, lite UI mode.

Tools & skills (representative): MCP integration; RAG / retrieval; filesystem, shell, sandboxed Python; browser automation (Playwright); research / web; vision; Whisper / STT; TTS; PDF; data analysis (CSV/Excel, charts); presentations (PPTX); spreadsheets (XLSX); website scaffold; computer-use / grounding (e.g. OmniParser ONNX); code workflows; engineering playbooks; LSP bridge (pyright, TypeScript language server).

Integrations: Telegram, Discord, Slack, WhatsApp; email, calendar; Obsidian; smart home (e.g. Hue); Twitter; Things 3 / file tasks; optional cloud LLM profiles when not local LLM only.

Operator & missions: mission graph, sub-agents, traces, best-of-N worktrees, forge promote/rollback API, slash commands (/restore, /undo), subtasks / total_steps progress.

Security & ops: API key auth, admin secret, safety envelope, endpoint guard, optional security scan, Docker, sandbox, structured logging, SECURITY / SECURITY_SETUP docs.

This Ollama artifact is the baked model slice; MCP, browser, RAG DB, and missions require the VIKI runtime.

Good use cases (this Ollama image)

• Self-hosted chat and offline assistant behavior (with a capable base)
• Private instruction tuning via system prompt (no extra pretraining)
• Q&A, summarization, coding help, documents, multimodal Q&A—when the base supports them
• Same VIKI “voice” as in-app when you point viki-evolved / model_name at viki-neural-forge

Build your own Neural Forge image

With the VIKI implementation checkout:

python scripts/build_viki_model.py

Public index: Documentation · Main README section: Build your VIKI model (in full repo docs).

License and commercial use

Commercial use follows the license of the underlying base weights. The Forge layer adds text and parameters only. VIKI tooling is Apache 2.0; you must comply with the base model vendor.

Topics / SEO keywords

ollama local-llm offline-llm self-hosted-ai private-ai ai-agent autonomous-agent tool-use mcp rag sovereign-ai local-ai python viki neural-forge viki-neural-forge modelfile qwen gemma llama reasoning multimodal air-gap open-source

Links

• Docs: https://github.com/Orythix/VIKI-public
  
• Model: https://ollama.com/orythix/viki-neural-forge