Skip to content
Start free

An AI analyst that cites evidence and asks first

Ask about your logs, metrics, traces and devices in plain language. Brain answers with links to the evidence, builds agents with you, and turns changes into proposals a person confirms, on the model you choose.

Start freeBook a demoRead the Brain docs
  • Bring your own model
  • Writes need confirmation
  • Auto-Fix never merges

How it works

From a question to a confirmed change

Brain is never more powerful than the person asking. It reads with your permissions, and it cannot approve anything.

  1. 01AskAsk in plain language, and @ mention a host, device or service to give Brain its context.
  2. 02ReadIts tools query logs, metrics, traces, problems and devices under your own permissions.
  3. 03ExplainAnswers cite the trace, the port and the change, with charts in the thread and links to each.
  4. 04ProposeA change comes back as a card with the exact target and the change. Confirm, edit or reject it.
  5. 05RouteDevice changes become NCM change requests and security responses become SOC proposals, each with its approvers.
  6. 06RecordEvery proposal and decision is written to a hash-chained ledger.

Workflows and agents

Build agents on a typed graph, with approvals the engine enforces

Wire triggers, AI agents with their models, tools, memory and MCP servers, approvals and actions on one canvas. Network configuration changes, Auto-Fix publishing and MCP servers allowed to write run only behind an approval: the engine refuses a graph that skips it.

  • Describe a workflow in a sentence and Brain drafts the graph for you to review and turn on
  • SOC triage, investigation and hunting agents ship as drafts you can read, edit and switch on
  • Ready-made playbooks install as ordinary workflows, with your credentials bound

Auto-Fix

From a production error to a pull request you review

Auto-Fix traces an error to the code, manifest or deploy behind it, has Brain write and verify a patch in its own worktree, and opens a pull request only after an approver says yes. It never merges.

  • Where errors come from

    Traces, logs, real user sessions, Kubernetes signals and failing CI jobs.

  • Checked before you see it

    Path rules, a secret scan and your own test commands run on the staged change.

  • Your code host

    GitHub, GitLab and Azure DevOps Services. Bitbucket is coming.

  • Confirmed in production

    A fix counts once its merge deploys and the error stays away for an hour.

Governance

You decide what AI can touch

Admins choose the models, decide which tools every agent may use, cap what agents spend, and can switch every model call off at once. Brain has no tool that changes those settings, reads a saved secret or approves anything.

  • One switch turns AI off everywhere: no model or embedding call is made
  • Auto-apply for small, undoable changes stays off until an admin allows it and the person picks it
  • Every Brain proposal and decision is on a hash-chained ledger an admin can verify

Better together

Brain works where the evidence already is

Problems, device changes and security alerts live in the same workspace, so an answer can cite all of them and a fix goes through the approvals each one already has.

Bring your own model

Which model providers work, and what each reports

Every AI feature runs on a connection your admins add. Pick one preferred model, a different one per feature, and a fallback for when a provider is rate limited or down.

ProviderConnect withModel listContext and limitsPrices for spend caps
AnthropicAPI keyLive from the providerReported per modelBuilt-in price table
OpenAIAPI keyLive, chat models onlyNot reportedSet by an admin
OpenAI-compatible serversBase URL, key optionalLive when the server lists them, or type the idWhen the server reports themRead from OpenRouter and Together, otherwise set by an admin
Anthropic-compatible gatewaysBase URL and API keyLive from the gatewayWhen the gateway reports themBuilt-in price table for Claude models

OpenAI-compatible servers include Ollama, vLLM, LM Studio, LiteLLM, OpenRouter, Groq and Together. An endpoint you host must be reachable from your workspace: private addresses are blocked unless an operator allows them. A model with no price counts as unpriced, not as free.

FAQ

Questions teams ask before they turn Brain on

Ask about AI and your data
What data does Brain send to the model provider?
Your message and what Brain's tools read to answer it, such as log lines, metric values, trace summaries and device names, capped in size for each tool call. Values that look like secrets, such as API keys, tokens and private keys, are replaced before the model sees them. Calls go to the provider and account your admins connect.
Does Brain change anything without asking?
Not by default. Every write is a proposal you confirm in a card. An admin can allow auto-apply for small, undoable changes, and then only a change you ask for in your own message applies at once, with Undo for 10 minutes. Device changes and security responses always go to their own approvers.
Can we use a model we host ourselves?
Yes. Connect any OpenAI-compatible endpoint, such as Ollama or vLLM, and choose it for every feature or just some. The endpoint must be reachable from your workspace; private addresses are blocked unless an operator allows them.
Can other AI assistants query Tracix over MCP?
Not yet. Today Brain and your agents can call MCP servers you add, but Tracix does not run an MCP server of its own.

Connect a model and ask your first question

Add an Anthropic, OpenAI or OpenAI-compatible credential, choose it in AI Settings, then open Brain from any page.

Start freeRead the guide