MindAlert
Alerting is noisy, so people mute it.
Alerting fatigue means the signal that matters gets lost in the signal that doesn't.
Proactive AI teammate
The AI teammate that watches every tool your team runs — and speaks up before you ask.
MindAlert collects from the tools you already run — logs, code, documents, any SQL database, the web, GitHub — works out what needs attention, and tells the right channel, with evidence. Nobody types a prompt.
Two failures happen at once
Evidence is scattered.
Diagnosis lives across logs, databases, code and chat; nobody reads all of it before deciding what broke.
Every "AI ops" tool still waits to be asked.
The expensive problem is the question nobody thought to ask at 3am.
One teammate, every tool you already run
Logs, documents, code, any SQL database, the web
Detect
Correlate into one incident
Investigate with evidence
Decide if it's worth a human
Message in chat
MindAlert collects from the tools a team already runs — log files, commands, HTTP endpoints, SQL against any database engine, MCP servers, documents and images, a browser for the internal tools with no API, GitHub. Detectors raise situations. Situations sharing a root cause are merged into one incident. An investigation runs, and a gate decides whether a human should be interrupted at all. Most of the time the answer is no — that is the point.
One visible tool to the model, however many capabilities sit behind it. It can even install a new one itself, from our own version-pinned tool catalog, the moment a customer asks for it.
Read-only by default. Runs on self-hosted models — your data never leaves your network.
Why we are not another agent framework
| Agent runtimes / assistants | MindAlert | |
|---|---|---|
| Trigger | You prompt it | It notices |
| Tool surface | Grows with every integration | One tool to the model, always — capabilities grow behind it |
| Claims | Plausible prose | Every claim cites a real log event |
| Wrong evidence | Passes through | Rejected at the gate |
| Volume | Every finding is a message | Gate decides; most never page you |
| Where the model runs | Vendor API | Self-hosted by default |
| Write access | Often implicit | Read-only; writes need human approval |
Hermes and the agent frameworks are runtimes — they are how you run a model. We are not competing with them; we replaced one in our own stack with two hundred lines, precisely so we are not locked to any of them.
Measured, not projected
305 → 0
Fabricated evidence citations, across 3 model families and 55 investigations.
11,344 → 39 → 2
Real 30-day log corpus: events → situations → messages sent to a human.
$0.14
Total inference cost for that 30-day month.
12 / 0 / 0
Prompt-injection attempts through chat: replies / credential leaks / unauthorised writes.
The corpus, cost and evidence numbers are reproducible from scripts/evaluate.py in our repository; the prompt-injection test is a separate scripted evaluation, also included.
Built for teams whose data cannot leave the building
- Read-only by default; a write becomes a proposal a named human approves in chat.
- No arbitrary shell access — command allowlists, timeouts, output caps.
- Inference stays on loopback/private addresses unless a customer explicitly opts out — nothing leaves the network by default.
- Permissive-licence-only dependencies, inventory enforced automatically on every change (341 acceptance tests, 673 unit/integration tests).
Annual subscription — support included
Priced in USD. No self-serve checkout — every deployment is configured to the customer's own tools, so pricing starts with a conversation.
Base platform — $10,000 / year
Includes support.
+ $2,000 / year per team
using MindAlert
+ $400 / year per custom integration
for any tool we build for you that isn't already in our public tool catalog. Tools already in the catalog (GitHub, any SQL database via usql, Google Drive, Calendar, mail, Microsoft 365, and more as we add them) are included at no extra cost.
Contact us
No purchase happens on this site — every plan is configured with you directly.
Talk to us
Whether you're an investor, a design partner, or just curious — we'd like to hear from you.