KT
Why is checkout latency spiking in production?
Private preview · In development
Incidents leave a trail.
Lino follows the evidence.
A self-hosted AI investigation partner that works inside Slack, gathers bounded read-only evidence from AWS and GitHub, and explains what it found — claim by claim.
Self-hosted · Read-only · Evidence-grounded
Lino
validating claims
01Slackquestion
02CloudWatchlogs
03AWSruntime
04GitHubchanges
Every conclusion stays connected to its source.
01 Read-only by construction
02 Evidence before inference
03 Self-hosted in your AWS account
04 Explicit gaps and confidence
01 The premise
Most AI answers ask you to trust the model.
Lino asks you to inspect the evidence.
01
Observe
Collect a bounded window of logs, runtime metadata, and relevant code changes.
02
Correlate
Let specialized agents compare signals without gaining permission to change your systems.
03
Verify
Review every material claim against stable evidence IDs before the report reaches Slack.
02 The Slack experience
A calm second pair of eyes, right where the incident starts.
Choose a question to preview how Lino separates observation, inference, and uncertainty.
L
I’ll trace the evidence across the production checkout path.
- Resolve target and permissions complete
- Inspect CloudWatch evidence complete
- Correlate AWS runtime and GitHub changes complete
- Validate claims complete
Degraded · high confidence
completed in 38s
Checkout latency rose after the inventory client began retrying synchronously.
Observation
P95 latency increased from 420ms to 2.8s at 10:31 UTC, isolated to checkout-api.
Inference
A retry configuration introduced in deploy v184 is amplifying inventory timeouts.
Hypothesis
The inventory dependency may be saturating before checkout retries begin.
needs upstream evidenceAI-generated analysis · read-only investigation · verify before acting
03 How Lino works
One question.
Five bounded stages.
Deterministic code controls identity, scope, limits, and rendering. Agents reason only inside those boundaries.
Resolve
Identify the exact app, account, region, and allowed evidence window.
Collect
Read redacted evidence from CloudWatch, Lambda or ECS, and GitHub.
Investigate
Send only relevant, bounded context to specialized agents running in parallel.
Review
Challenge unsupported conclusions, contradictions, and missing evidence.
Report
Return a cited explanation, explicit gaps, and one practical human next step.
04 Evidence contract
No floating conclusions.
Every observation points to evidence. Every inference names its support. Every hypothesis is visibly labeled — and collection failures stay visible as gaps.
- 01 Stable evidence IDs
- 02 Claim-level citations
- 03 Explicit contradictions
- 04 Calibrated confidence
05 Architecture
Small control plane.
Strict trust boundary.
Lino lives in your AWS account and assumes temporary, read-only roles in explicitly configured target accounts.
Interface
#SlackDM + @Lino
Your control account
Ingressacknowledge
SQSqueue + DLQ
Workeragent team
DynamoDBruns + config
STS temporary credentials
Independent target accounts
AWS account Aread-only role
AWS account Bread-only role
GitHubread-only token
Read-only by construction. Lino has no customer-system mutation tools. Denied access fails closed and becomes an evidence gap — never an invented fact.
Bounded access
Allowlisted accounts, regions, applications, resources, and evidence windows.
Visible failure
Denied, missing, timed-out, and truncated evidence stays visible in the report.
Recoverable runs
Idempotency, leases, checkpoints, cancellation, bounded retries, and a DLQ.
06 Product clarity
Questions worth answering before you trust an incident agent.
Clear boundaries are a product feature, not fine print.
01 Is Lino production-ready today?
Not yet. Lino is in active development. The first release is gated by reproducible deployment, failure testing, evaluation, security review, and a synthetic demonstration.
02 Can Lino change production systems?
No. The product boundary is read-only. Lino collects evidence from explicitly allowed operational sources, then returns a cited report and one human next step.
03 Where does operational data go?
The control plane is self-hosted in your AWS account. Evidence is bounded and redacted before model context. The configured model provider is an explicit data-processing choice and invalid provider configuration fails closed.
04 What will the first release investigate?
The first release focuses on CloudWatch Logs, Lambda and ECS runtime metadata, and GitHub deployment context, all operated through Slack direct messages and @Lino mentions.
05 When will Lino be available?
Availability follows the first-release gates: safe installation, verified read-only access, reliable recovery behavior, evaluation targets, and a reproducible demonstration. This page will reflect the release status as those gates are completed.
Private preview · Active development
Self-hosted.
Read-only.
Built to be trusted.
Lino is not an auto-remediation bot. It is a careful investigation system that keeps conclusions connected to evidence and decisions in human hands.