Brutlabs LLC · Network automation & AI
Foundational network automation, built for agentic AI operations
Brutlabs builds the source of truth, config pipelines and telemetry your network is missing, or uses your existing automation infrastructure as semantic context to deliver AI driven triage, root cause analysis and human-approved remediation for network problems.
Root cause
Change #8842 shipped a template revision that dropped the jumbo-frame MTU on bb-core-2 eth3/1. Oversized BGP updates are being dropped, so the session resets.
Proposed fix
interface eth3/1
mtu 9214
Rollback: revert #8842 template block. Blast radius: 1 interface, 1 device.
Your network already contains the answer. Nothing reads it together.
Every signal needed to diagnose a typical fault exists somewhere in your estate. It is spread across different systems that do not talk, and correlating them is a manual job that happens at the worst possible hour.
No source of truth
Reality lives in spreadsheets, tribal knowledge and stale diagrams. Nothing an automation, or an agent, can safely act on.
Undocumented configuration
Running state has drifted from intent, and nobody is certain what is actually deployed where.
Manual provisioning
Circuits and configs are hand-built, slow, and different depending on which engineer built them.
Alert fatigue
Thousands of low-signal alerts bury the one that mattered. Everything is a P1, so nothing is.
Config drift
Intended and running state diverge silently until something breaks at two in the morning.
Siloed monitoring
Telemetry, logs and metrics sit in separate tools that no human can correlate quickly enough to matter.
What it does during an incident
Not a chatbot on top of your alerts. An agent that decides what to look at next, gathers the evidence itself, and keeps going until it can name a cause.
Reads all four layers
Source of truth, config change history, telemetry and logs — queried directly at the raw level, not through a pre-digested alert.
Correlates across them
A latency shoulder, a flapping session and a config deploy six minutes earlier become one story about one interface.
Compares state against intent
What the device is actually running against what it should be running. The highest-yield question in network diagnosis, and it needs a source of truth to ask.
Recalls past incidents
Retrieval over your own incident history and runbooks: has this template, this device, this signature caused trouble before?
Proposes a fix and a rollback
Delivered into the same pipeline your engineers use, with the blast radius stated and the evidence attached.
Stops for approval
Human-in-the-loop by default. Autonomy widens per action, on evidence, when you decide it has earned it.
The foundation that makes an agent worth having
An AI agent is only as trustworthy as the data layers under it. We build the layers that make its reasoning sound.
Source of truth
NetBox or Nautobot built properly: a data model that fits your network, real discovery and reconciliation, and an API contract every pipeline reads from.
Explore →02Config automation & CI/CD
Ansible, Nornir, Terraform and Jinja2 behind a pipeline that validates before it deploys and records every change in a form the agent can trace.
Explore →03Telemetry & observability
High-fidelity telemetry — gNMI where the platform supports it — plus Prometheus, Grafana and centralised logging, tuned so signals correlate instead of competing for attention.
Explore →04Zero-touch provisioning
Devices and sites that configure themselves from the source of truth, then verify their own day-0 state before anyone calls it done.
Explore →05Intelligent alerting
Alerts enriched with source-of-truth context — device role, site and intended state — so a page reflects what is actually wrong on the network, not which counter crossed a line.
Explore →06Auto-remediation
Codified runbooks the agent can execute against known conditions, with approval gates, blast-radius limits and a full audit trail.
Explore →What changes once the foundation and the agent are in place
Representative outcomes from a mid-size enterprise engagement.
A staged path, not a science project
Four stages, in this order, because each one depends on the last. A comprehensive roadmap for a robust agentic automation framework.
Standardise
Structured data and a single source of truth. Nothing else is safe to build on.
Implement
Version-controlled config management and automated provisioning, validated before it deploys.
Observe
High-fidelity telemetry, centralised logs and alerting that earns the page it sends.
Delegate
The agent reads all of it, diagnoses faults, and remediates known conditions under your guardrails.
Open-source first, and tool-agnostic where it counts
We build on the tools your engineers already know, so what we hand over is maintainable by the people who have to live with it.
No vendor lock-in. No multi-year engagements. You own what we build.
Inventory, IPAM and intended state. Everything downstream — every playbook and every agent query — reads from here.
Turning intended state into deployed reality, with validation before the change lands and a change record the agent can trace back to.
State at a resolution that shows the fault, and logs in one place — so an agent can query raw signal rather than second-hand alerts.
The reasoning tier. Orchestration, tool access, memory of past incidents, and the guardrails that keep it inside its lane.
Common questions
What does the Brutlabs debugging agent actually do?
It connects to your source of truth, config pipeline, telemetry store and logs, and during an incident it queries all four in a reasoning loop — gathering evidence, correlating signals, comparing running state against intent and checking past incidents — until it can state a root cause and propose a fix with a rollback plan. It stops for human approval before changing anything.
Do we need to have automation already to work with Brutlabs?
No. Most clients start with gaps in the foundation, and building that foundation is the bulk of the work. We assess where you stand, build the missing layers alongside your team, and connect the agent once there is enough trustworthy data for it to reason over.
Is the agent allowed to change our network?
Only within limits you set, and never by default. Every action starts in propose-only mode. Approval gates, blast-radius policy and full audit logging are on from day one, and each action earns wider autonomy on evidence rather than on trust.
How is this different from AIOps products we have been pitched?
Most AIOps tooling sits on top of alerts and clusters them. The agent works from the layer below — raw telemetry, config diffs and intended state — and it acts in a loop rather than producing a single classification. It also comes with the engineering to make that data exist, which is usually the missing half.
What does an engagement look like?
A 30-minute readiness consultation, then a written assessment. From there a typical build runs in phases: source of truth, config pipeline and change telemetry, observability, then agent enablement. We work with your team rather than around them, and you own everything we build.
Book a 30-minute automation readiness consultation
In 30 minutes, we’ll evaluate your infrastructure maturity, identify operational risk areas, and highlight high-impact automation opportunities.
No scripts. No invasive discovery. Just clarity.