Keel is AI-native. Ask it anything about your logs in plain English, get automatic root-cause analysis the moment an incident fires, and connect your own AI agents over MCP — all running against the AI model you choose, with your log data staying in your own environment.
Most tools add a chatbot on the side. In Keel, the AI is the interface: you ask questions, it investigates incidents for you, and it opens up to your own agents. Every piece runs on a model you control.
Ask in plain English and Keel answers — a result set, a chart, or a summary. Save what you ask and pin it to a dashboard.
When a detection raises an incident, Keel investigates — correlating logs and reading your runbooks — and delivers a root cause in plain English.
A chat assistant that investigates in plain English — search logs, check the appliance, look up incidents — and can take actions, like adding a source or drafting a detection, that it asks you to confirm first.
Connect your own AI agents — Claude, for example — directly to your logs and Keel's intelligence. Read-only and scoped.
Type a question the way you'd say it — "show 5xx errors from the payment service over the last two hours" — and Keel figures out the query and answers. No query language to learn.
When a detection fires, Keel doesn't just page you — it investigates first. It correlates the surrounding logs across services, reads the runbooks you've uploaded, and writes up a root cause in plain English, delivered where your team already works.
Keel ships a Model Context Protocol (MCP) server, so the agents your team already uses — Claude, for example — can investigate your logs and Keel's intelligence directly, in a conversation. It runs inside your network with the rest of Keel, so your agents reach your logs without your logs reaching a vendor cloud.
MCP is live in the product; dedicated MCP documentation is on the way. Ask us for setup help during your evaluation and we'll walk your team through connecting an agent.
Keel ships no bundled model. You point every AI feature at the model you trust — and you can change your mind whenever your policy does.
Run a local model (for example Ollama) on your own network via any OpenAI-compatible endpoint — for fully air-gapped sites where nothing leaves at all.
Use your own Anthropic, OpenAI, or Google Gemini key — or any OpenAI-compatible endpoint (Azure, Groq, vLLM, Mistral…) — under your own account and that provider's terms.
Configure a primary and a secondary backend that takes over automatically if the primary is unreachable. No model configured? Keel is still a complete conventional log platform.
This is the distinction that wins the room: your log data and the AI request are two different things — and Keel keeps you in control of both.
The data plane stays home. All ingest, storage, search, and dashboards run on infrastructure you own. No log content, query text, or infrastructure detail ever goes to Infranexis, and there's no phone-home to operate.
The AI request is under your control. When you trigger an AI feature, only the context for that request goes to the model endpoint you configured — your provider, your key. Choose a local model and nothing leaves your network at all. It happens only on demand, when you ask or an incident fires.
Bring a question you'd normally spend an afternoon on. We'll show you Keel answering it — on the model you choose, inside your own environment.
Book a walkthrough Try the live demo