Local-LLM AI agents — private, predictable, enterprise-grade
Run AI agents on models that live on your own hardware. No data leaving your network, no per-call API bill, and the option to reach for a cloud model whenever you want.
Why run agents on local LLMs?
- Privacy — prompts and data never leave your machine; nothing is sent to a third-party model provider.
- Predictable cost — no per-token API fees; run as many tasks as your hardware allows.
- Offline-capable — automations keep working without an internet connection to a model API.
- No lock-in — swap models freely as open models improve.
AIK runs local models through Ollama — pick any installed model in Settings and your agents use it. All models (local and cloud) are chosen from one picker; there's no code to change.
Cloud flexibility when it earns its place
Local-first doesn't mean local-only. Choose OpenAI, Anthropic Claude, Google Gemini or Groq per use case, and run everything else locally. Use a big cloud model for the one hard task and keep the rest private and free.
Real tools, not just chat
Agents are only useful when they can do things. AIK ships built-in tool servers (files, email, databases, web browsing, documents and more) so agents read inboxes, query databases, browse sites and process documents out of the box — extensible via the open MCP standard.
Engineered to run in production
Local and self-hosted only pays off if it's dependable. AIK is built for unattended operation:
- Failure isolation — every tool server is its own process; one failure never brings down the platform.
- Dependency validation & health checks — confirm each tool server actually starts and responds, so a bad update is caught immediately.
- Per-server timeouts & graceful degradation — a slow tool times out and reports "unavailable" instead of hanging.
- Safe rollback — updates snapshot the app and database first; one click restores the previous working version.
- Signed updates & error history — updates are verified before they apply, and failures are recorded to a tamper-evident log.