Forge connects to the tools you already run, learns how your systems fit together and works every incident from the first alert to a verified fix.
A live map of every service, owner and dependency
Forge builds its picture of production from the signals you already emit. Each connection carries a confidence score and a history you can rewind to any moment.
- Built from traces, network flows and deploy config
- Owners and on-call pulled from your existing tools
- Rewind the map to before any deploy
- Finds services nobody remembers running
One incident per real problem, with the cause attached
Alerts, log bursts, failed traces and deploys are placed on the map and grouped by topology, timing and shape. Every group shows its evidence and its confidence.
- Works with the alerts you already have
- Probable cause named with a confidence score
- Grouping mistakes are tracked and published
- Pages the owning team first
A queue ordered by what it costs you
Incidents and watch findings share one queue, ranked by users affected, revenue at risk and error budget burned. When something new arrives, the order updates itself.
- Impact estimated from your real traffic
- Incidents and slow trends in one place
- Act now, This week and Backlog lanes
- Weights you can tune per service
Fixes that arrive with their proof
Forge drafts the pull request, runs your tests, replays the incident traffic and attaches a rollback plan. A person approves, and a canary decides whether it reaches everyone.
- Pull requests that follow your conventions
- Tests and replayed traffic before review
- Canary promotion rules your team owns
- Automatic rollback when numbers slip
Seven agents that never go off shift.
Each watch runs on its own schedule, files its findings into the same priority queue and drafts a fix when one is needed. You can also describe your own in plain English.
Latency watch
Every 60 seconds
p95 and p99 regressions on critical paths.
- Endpoints watched
- 486
- Regressions this week
- 3
- Median time to catch
- 74s
Recommendation
Keep search-api warm on weekday mornings
p95 rises by 120ms at 07:00 UTC on weekdays because of cold starts after the nightly scale-in. Keeping 4 instances warm between 06:30 and 09:00 removes it for about $190 a month.
Fix drafted and waiting for review
Connects to the tools you already run.
No new agents to install and no dashboards to rebuild. Forge reads from your existing stack, read-only to start.
Observability
- Datadog
- Grafana
- Prometheus
- New Relic
- Honeycomb
- OpenTelemetry
- Sentry
- Elastic
Cloud and runtime
- AWS
- Google Cloud
- Azure
- Kubernetes
- Terraform
- Cloudflare
Code and delivery
- GitHub
- GitLab
- Bitbucket
- Argo CD
- Buildkite
- CircleCI
Incidents and chat
- PagerDuty
- Opsgenie
- Slack
- Microsoft Teams
- Jira
- Linear
You approve every change.
Forge is built to be trusted with production, which means it is built to be checked. These are defaults, not settings you have to find.
Read-only by default
Every new connection starts read-only. Forge can watch and reason long before it is allowed to propose anything.
Scoped permissions
Write access is granted per repository and per environment, and can be withdrawn in one click.
Proof before review
Tests, replayed incident traffic and a rollback plan are attached before a person is ever asked to look.
A full audit trail
Every query, suggestion and change is logged with its evidence and the person who approved it.
Run it where your data lives.
Forge Cloud
Fully managed, running in the US or the EU, ready in an afternoon.
- Data residency in the US or EU
- Encrypted at rest and in transit
Dedicated
A single-tenant Forge with its own network, keys and upgrade schedule.
- Customer managed encryption keys
- Private networking to your cloud
Your VPC
Forge runs inside your own cloud account. Telemetry and code never leave it.
- No outbound data to Forge
- Built for HIPAA and regulated teams
Give production a steady hand.
See Forge map your services, group your alerts and draft its first verified fix, on your own stack, in a 30 minute walkthrough.