Pre-GA · recruiting design partners
Incident response that leaves a record your auditor can verify.
On-call and incident response for regulated enterprises. Every state change, agent recommendation, human approval and executed action hashes into a per-tenant chain. Alter one record and the chain stops verifying.
The record beside this is live. The hashes are computed in your browser from the bytes on the page. Change who approved the rollback and watch it break.
✓ Chain verified
5 entries · SHA-256 · tenant tnt_4f2a
Chain verified. Every entry hashes to the value its successor recorded.
- 0041state transitionP1 declared on checkout-api
- incident
- INC-4471
- service
- checkout-api
- state
- declared
- severity
- sev_1
- commander
- a.mensah
prev 00000000…0000hash ac2721c8…457c - 0042agent decisionRemediation agent proposes a rollback
- agent
- remediation
- retrieval_set
- rs_9c31e0
- proposes
- deploy_rollback
- confidence
- 0.87
- blast_radius
- service
prev ac2721c8…457chash c5198a48…cdc7 - 0043approvalSRE on-call approves the action
- action
- a_88c1
- approver
- s.okafor
- policy
- tier0_deploy_rollback
- basis
- blast_radius:service
prev c5198a48…cdc7hash 2d1d4cfb…cd50 - 0044actionRollback executes and succeeds
- action
- a_88c1
- runbook
- rb_checkout_rollback@2.1.0
- state
- succeeded
- rollback_available
- true
- duration_ms
- 41216
prev 2d1d4cfb…cd50hash 63f9d496…5064 - 0045evidenceFiled as SOC 2 evidence
- artifact
- ev_2f70a4
- framework
- soc2
- control
- CC7.4
- derived_from
- action:a_88c1
prev 63f9d496…5064hash fa7d4a1b…43cc
Check entry 43 by hand
printf '%s' "c5198a488169df8d3c2c0b2f32957275ac4a40c915197dfdc0e5cd489235cdc7|{\"action\":\"a_88c1\",\"approver\":\"s.okafor\",\"policy\":\"tier0_deploy_rollback\",\"basis\":\"blast_radius:service\"}|2026-03-11T14:06:30Z|tnt_4f2a/2026-03" | shasum -a 256Worked example. The operational details are illustrative. The hashes are not: they are SHA-256 over exactly the records shown, in the entry format of Spec §5.8.
The problem
The AI you are being sold cannot go in the audit file.
Every operations vendor shipped an assistant in the last two years. For a regulated buyer the question was never whether the model is any good. It is whether you can show an assessor what the model saw, what it recommended, who approved it, and what changed in production as a result.
Most tools cannot answer that, because the answer was never recorded. Their AI was built as a productivity feature, so provenance was never part of the data model. Adding it now means breaking the integrations their existing customers depend on. That is the opening StaidKit is built for.
Scope
What it replaces, and what it deliberately does not.
Replaces
- On-call and paging.PagerDuty, Opsgenie
- AI-assisted incident response.Incident.io, Rootly, FireHydrant
Integrates with, and does not replace
- Observability, read-only.Datadog intake implemented. Others specified.
- Service management, bidirectional.ServiceNow synchronisation implemented.
- Compliance automation, evidence push.Vanta and Drata hold configuration evidence. We produce action evidence. Specified.
State of build
What is running, and what is still a design.
Marking our own gaps is unusual on a page like this. It is also the only way a buyer who is shown an AI claim every week can tell which parts of this one are checkable today.
| Capability | Status | Where |
|---|---|---|
Incident and crisis operations Declare, assign a commander, run the timeline, publish status updates. | Implemented | crisis-service |
Telemetry intake and live feed Event ingestion from connected monitoring, streamed to the operations surface. | Implemented | telemetry-service, websocket-service |
Failure prediction Ensemble models over ingested telemetry, with a separate retraining platform. | Implemented | prediction-engine, ml-platform |
Observability and service management integrations Datadog intake and ServiceNow synchronisation. | Implemented | integrations/datadog, integrations/servicenow |
Asset inventory and dependencies Assets and CMDB synchronisation are running. Graph traversal and freshness reconciliation are not. | Partial | asset-service |
On-call schedules and escalation Schedules, rotations, escalation policies, automatic escalation and SLA-breach checks, with their own migrations and tests. | Implemented | crisis-service |
Compliance controls and evidence SOC 2, ISO 27001, HIPAA and DORA control models exist. Automated evidence collection does not. | Partial | compliance-service |
Cryptographic action ledger The hash-chained record every other claim on this page depends on. | Specified | not built |
Typed runbook language Statically verified rollback, and blast radius computed from the service graph. | Specified | not built |
Scribe and remediation agents Summarisation and action proposals over per-tenant operational memory. | Specified | not built |
The bet
Four decisions the incumbents cannot retrofit.
Every individual feature in this category already exists somewhere. What does not exist is a system where provenance is the foundation rather than a later addition, because adding it later breaks the installed base.
A per-tenant operational memory
PartialIncidents, runbooks, dependencies, approvals and evidence in one graph that the agents read from and the runbook compiler computes against. No tenant's data trains anything for another tenant.
Spec §8
A hash-chained action ledger
SpecifiedEvery state change, agent decision, approval and execution appended to a per-tenant chain, anchored to a log you control. An auditor verifies it without trusting us.
Spec §9
A typed runbook language
SpecifiedEvery step declares its preconditions, blast radius, cost and rollback. Programs that fail static verification do not run. A model can emit this. A model cannot be trusted to emit shell.
Spec §10
Approval before action, by default
SpecifiedAgents propose. People approve. Autonomous execution is not in the first release, because it is not a posture a regulated operator can defend today.
Spec §7.3
How the ledger, the graph and the runbook language fit together
Where we are
Pre-general-availability, and saying so.
Eleven backend services run today, covering incidents, telemetry, prediction, assets, authentication and compliance models, with Datadog and ServiceNow connected. That is a real foundation, and it is not the product yet.
The four decisions above are what turn it into the product, and three of them are not built. We are not going to write them in the present tense to make this page read better.
There are no customers yet, so there are no logos, no case studies and no retention figures here. There is no SOC 2 report for StaidKit itself yet either. The first audit is part of the road to general availability.
What we want now is five design partners who will shape the first release and hold us to this standard of disclosure.
Design partner programme
Five partners, and a first release shaped around what you actually have to prove.
Two in financial services, two in health technology, one critical-infrastructure platform. If you run on-call today and assemble your audit evidence by hand, this is built for you.