Transcript
The InOrbit tour
A short narrated tour of InOrbit: monitor, respond, understand with Atlas, verify, and how to get started.
Narrated by an AI voice (Bella, from ElevenLabs). The script was written for InOrbit and approved by Nevio Vesić (owner) on 2026-10-08. Version 9, approved 2026-10-08.
0:00Understand, detect, prove
InOrbit is the engineering platform that helps you understand your system, detect what breaks, and prove what fixes it.
On screen
- Understand your system.
- Detect what breaks.
- Prove what fixes it.
- Monitor, respond, understand and verify, in one loop, with the evidence kept.
0:07Monitorlive
It starts with monitors. They check your endpoints on a schedule, from our cloud or from an agent inside your network.
And every single run becomes evidence: what was checked, where, when, and what actually happened.
On screen
- Monitor · live
- checkout-api · GET /healthz · every 30 s · example data
- 14:02:00 2.9 s 503 from our cloud · example data
- 14:02:00 3.1 s 503 from your agent · example data
- Every run becomes evidence.
0:21Respondpreview
When something breaks, an incident opens on the monitor that saw it, and you acknowledge it right from your phone.
Its impact is measured from real runs. Not estimated. Not guessed.
On screen
- Respond · preview
- The monitor fails from both places · example data
- Opened from the monitor page, responders named · example data
- Acknowledged from the lock screen · example data
- 29 min · 4.2% of checks failed · p95 3.4 s · example data
- impact measured from 58 monitor runs, not estimated · example data
- Today a person opens the incident. Alerts that open it by themselves come next.
0:34Every sensor, one evidence model
Monitors are just one sensor. The agent, Envoy, Kubernetes, your compilers, GitHub, the browser extension and your team all report in.
Every source declares what kind of evidence it gives, and Atlas decides what that evidence can prove.
We tried it on our own platform. Nobody told it anything, and it mapped how the system is really built: 4,796 facts, each one linked to its source.
On screen
- HTTP & gRPC monitors · live
- Agent checks in your network · preview
- Envoy access logs · in development
- Trails browser recordings · preview
- eBPF capture, optional · coming
- Envoy routes · in development
- Kubernetes manifests · in development
- Kubernetes API, read-only · in development
- Network policies · in development
- Cargo metadata · in development
- rust-analyzer · coming
- go/types · coming
- TypeScript compiler · coming
- Java JDT · coming
- Python AST · coming
- tree-sitter, the rest · coming
- GitHub pull requests · live
- incident.io · in development
- Grafana · Loki · coming
- Datadog · coming
- PagerDuty · coming
- Cloud providers · coming
- Reviews of PRDs, ADRs, RFCs · live
- Acknowledged from the phone · preview
- Owner's answers · coming
- Docs read by a model · coming
- Every item keeps its method, its source and how it can fail
- Atlas decides what each kind can prove.
- A model's reading is never proof on its own.
- real run · 2026-10-08
- We tried it on our own platform.
- Nobody told it anything. The agent read the system and mapped how it is really built.
- 4,796 facts, each linked to its source
- 462 parts of the system found
- 32 running services mapped
- What it worked out by itself:
- ✓ Which service answers each web address
- ✓ Which services are allowed to reach the internet
- Every fact can be traced back to where it came from. Nothing is a guess.
1:00Understand with Atlasin development
Atlas, now in development, turns that evidence into a living map. Ask where a request really goes, and it traces it hop by hop, from Envoy through Kubernetes down to the process, with proof for every step. Down in the kernel, the agent will see the traffic itself: packets, connections, DNS and drops. Until then, Atlas names that hop as not seen yet.
And when the running system breaks an architecture decision, it shows both sides, and when each one became true.
On screen
- Understand · Atlas · in development
- real topology · example trace
- GET /v1/me, traced from the edge to the code and back through the engine load balancer · example data
- contradicted
- paging makes no calls outside our network · example data
- allows paging and nine other workloads egress to any public host on 443/TCP
- Manifest and live cluster agree. Both sides shown; Atlas doesn't pick one.
1:33Ask your systemin development
And because every claim has its proof, you can simply ask. Anyone in engineering, or a sales engineer live on a customer call, gets an answer with its sources, and the gaps named honestly.
On screen
- Atlas · in development · in development
- What breaks for customers if the ledger goes down? · example data
- Checkout can't complete payments. · example data
- It calls the ledger on every order. · example data
- ADR-0017 caps retries at 3, so customers see an error in about 2 s. · example data
- Refunds queue and recover later. · example data
- Not known yet: whether the nightly export needs it. · example data
- Every answer cites its evidence. What isn't known is said, not guessed.
1:45Verify, a real runpreview
When a change ships, the verifier measures its claims before and after it rolls.
Our very first run failed, and the record says so. No pretending, and no confidence scores.
On screen
- Verify · preview
- real run · 2026-10-04
- mise run verify:before 251
- 44 passed, 0 failed
- mise run verify:after 251
- 43 passed, 1 failed
- VERDICT: FAIL
- Posted on pull request #251.
- Three claims held, one check broke, and the record says so. A verdict only covers what was tested.
1:57Install and connect
Getting started takes minutes. Install the command line, sign in, and add the agent with one command. It connects out only, so nothing opens inbound.
Then connect the tools you already use, like GitHub, your incident tool and Grafana, and your AI assistant through MCP: Claude today, others next. Each connection asks only for the access it needs, and becomes another source of evidence.
On screen
- brew install inorbithr/tap/iohr
- iohr login
- iohr ext install agent
- outbound only · nothing opens inbound
- command line · live · live
- agent · preview · preview
- the agent is on every plan, Free included
- GitHub · live
- Kubernetes, via the agent · in development
- incident.io · in development
- Claude, via MCP · preview
- Other AI assistants, MCP · coming
- Grafana · coming
- Loki · coming
- Datadog · coming
- PagerDuty · coming
2:24One loop
Monitor, respond, understand, verify. One loop, where the evidence feeds the next turn.
Early access is open for partners. Request access at inorbit.hr.
On screen
- Monitor · live
- Respond · preview
- Understand · in development
- Verify · preview
- the evidence feeds the next turn
- Every system you run gets its own loop.
- early access · partners
Narrated by an AI voice (Bella, ElevenLabs).