01
Scope the incident
Choose the exact Slack channels, time window, threads, and reviewer. Nothing is searched in the background.
Evidence-first incident review
The incident is over.
The truth isn’t ready.
OnRecord turns fragmented Slack conversations into a source-linked postmortem your team can verify, revise, and stand behind.
Source linkedHuman approvedHistory preserved
OR
INC-DEMO-1042EU checkout outage
Needs reviewSUPPORTED FINDING
An unauthorized WAF rule blocked checkout requests at the edge.
3 sourcesDirectly observed
OPEN QUESTION
How was the authenticated session acquired?
Slack conversationsEvidence graphApproved record
The dangerous part isn’t the blank page
It’s the confident sentence nobody can defend.
During an incident, observations sit beside guesses. Decisions are buried between status updates. Contradictions arrive out of order.
A conventional AI summary smooths that mess into fluent prose. OnRecord does the opposite: it preserves the seams, so a reviewer can see why every conclusion exists.
The actual review workspace
This is product code, not a product rendering.
The window beside this copy loads the same React application used by the authenticated review console. The public demo swaps only the API boundary for validated synthetic state.
Real console components Synthetic evidence External effects blocked
Use the actual interface 90-second product walkthrough
Watch one incident travel from Slack to an approved Confluence record.
Recorded in a connected test environment with synthetic incident data. Processing and review sequences are visibly accelerated; the approval and publication path is real.
Connected test environment01:30 · Slack to Confluence
From signal to record
A disciplined path through messy evidence.
No autonomous diagnosis. No one-prompt postmortem. No silent upgrade from “maybe” to “fact”.
02
Build the evidence graph
Messages become cited timeline events, claims, contradictions, and questions—not a loose prompt transcript.
03
Review every conclusion
Keep, edit, exclude, or reclassify each statement while the source evidence stays beside it.
04
Approve one record
Only a human-approved immutable revision can be published to Confluence or Notion.
Show the work
Four product boundaries.
Each one is inspectable.
“Implemented” means code and tests exist. It does not mean a customer environment has already been provisioned.
01
Slack intake
Scope what the system is allowed to see.
A signed Slack shortcut opens a bounded scoping flow: time window, one primary public channel, up to four additional public channels, optional thread anchors, and a named reviewer.
Implemented · requires connected Slack02
Evidence pipeline
Create the record before writing the story.
Collection is checkpointed. Analysis persists cited timeline events, claims, contradictions, questions, source coverage, model metadata, and token usage before report generation begins.
Implemented · synthetic demo data03
Human review
Use the same interface reviewers use.
The public demo is a production build of the real review console—not a marketing recreation. Only its API boundary is replaced with validated, in-memory synthetic incident data.
Live in this demo04
Controlled publication
Publish exactly what was approved.
Approval locks one immutable revision and queues retry-safe publication through configured Confluence or Notion adapters, followed by a final Slack notification.
Implemented · external effects disabled hereCapability ledger
What exists, what runs here, and what comes next.
Implemented
Code and focused automated tests exist in the product repository.
- Signed Slack trigger and incident-scoping modal
- Multi-channel collection, thread expansion, and coverage manifest
- Structured claims, timeline, contradictions, and open questions
- Evidence-constrained report generation and citation validation
- Authenticated review console with immutable revision history
- Human approval and Confluence or Notion publication adapters
Proven here
The production review UI runs against constrained synthetic state.
- Inspect claims, timeline events, evidence, and source coverage
- Keep, edit, exclude, or add evidence-linked statements
- Answer open questions and acknowledge contradictions
- Save a revision, approve it, and inspect the read-only result
Customer setup
Implemented paths that still require a provisioned customer environment.
- Slack app installation and approved workspace scopes
- AWS, PostgreSQL, Cognito, networking, and runtime secrets
- A destination Confluence space or private Notion data source
- Operational ownership for retention, alarms, and deployment promotion
Future work
Explicit roadmap—not represented as available product functionality.
- Slack OAuth installation and token lifecycle management
- GitHub App evidence collection
- Enforced retention-deletion jobs
- Action-item creation and semantic quality benchmarking
Capability status reflects the repository as of July 2026. The public demo never connects to customer systems or publishes externally.
Live system journey
Watch one event become an approved record.
The moving packet follows the implemented production path. It pauses at the human gate because infrastructure can prepare a record, but it cannot approve one.
Tracing event · incident.review.requestedSynthetic animation
01 · Intake
Accept the event durably
Slack event
Raw signed HTTP request
API Gateway
Public webhook boundary
Ingress Lambda
HMAC + replay verification
SQS FIFO
Durable, deduplicated command
Worker Lambda
Idempotent incident start
Step Functions
Deterministic orchestration
02 · Evidence pipeline
Collect, structure, and draft
Collector Lambda
Bounded Slack history pages
Slack Web API
Approved channels + threads
Evidence store
Atomic PostgreSQL checkpoints
Analysis Lambda
OpenAI structured extraction
Report Lambda
Source-linked draft + validation
Notify Lambda
Content-free review-ready link
03 · Human gate + publication
Review, approve, then publish
Review console
S3 origin through CloudFront
Amazon Cognito
Authorization code + PKCE
Review API Lambda
JWT + membership authorization
Approved revision
PostgreSQL transaction + outbox
EventBridge
Bounded publication schedule
Publisher Lambda
Leased, retry-safe side effects
Final destinations
Confluence or Notion + Slack
External boundary AWS runtime Durable record Human authority
Secrets Manager supplies scoped runtime credentials to the relevant Lambdas and is intentionally shown outside the event path: secrets configure execution; incident payloads do not travel through it.
The implementation stack
Technology chosen for durable evidence, not demo theatre.
These technologies are present in the repository. Customer infrastructure still requires explicit provisioning.
01
Experience
React 19
TypeScript
TanStack Query
Radix UI- Amazon Cognito
02
Evidence + AI
OpenAI Responses API
Zod schemas- PostgreSQL
AWS Secrets Manager
03
Workflow
- API Gateway
- AWS Lambda
- SQS FIFO + DLQ
- Step Functions
- EventBridge
04
Delivery + release
- Amazon S3
- CloudFront
Terraform
GitHub Actions- Confluence
- Notion
Brand marks identify the technologies used; they do not imply vendor endorsement. AWS service artwork is from the current AWS-approved architecture icon set.
Trust is a product behaviour
The model drafts.
It does not decide.
Evidence before prose
The draft is generated from structured, cited claims—not directly from raw conversation.
Uncertainty stays visible
Hypotheses, disputes, partial coverage, and unanswered questions cannot be polished away.
Approval means something
The model cannot confirm a cause, approve a revision, or create an external effect.
Built with tenant-scoped authorization, immutable revisions, bounded model access, retry-safe publication, and source-content-free operational logs.
Human approval requiredSynthetic incident. Production interface.
See what changes when every conclusion has to show its work.
Review the evidence, challenge the draft, and put one revision on the record.
Enter the demo