Runtime surprise
The agent starts failing a tool call at 02:00; support finds out from customers.
boardRisk flags fire, the alert lands in Slack, an incident opens with a timeline.
NewTrust Score, public Trust Center, proof chain, and Annex IV files are live.Trust Center launch kit live.See what changed
ingestOKgatearmedchainverifiedshift24/7
Run AI like you run production.
AegisRail is the watchfloor for production AI: traces in, regressions held at the gate, every system owned, evidence sealed. The board below is the product.
Open a workspace
pnpm add @aegisrail/sdkTrace feedlive
tr_2af91c7support-agent · refund exceptionflag
ev_3c12a4brefund-policy-eval · 42/44 passpass
tr_2af91d2onboarding-agent · tool failurefail
bd_71e0fc9evidence bundle · CC7.2 sealedseal
in_55a8c12IN-118 · drift on intent classopen
tr_2af9214support-agent · pii redactedok
ap_8d2f01aC-15 access review · signedok
tf_413799cannex iv file · support-copilotfile
Release gateholding
refund-policy-eval · v342✓ · 2✗
- case_28pass
- case_29pass
- case_30pass
- case_31fail
- case_07fail
- case_32pass
Release held
GET /api/evals/:id/gate → fail
pipeline exit 1 · deploy blocked
Trust · evidencesealed
92
Trust Score
grade A · 8 systems
- eval verdict recorded4f19…c2aa
- IN-118 resolved77b0…91d4
- bundle sealed9c4e2af3b1d
proof chain · public verify endpoint
board data: synthetic
RSK · failure modes
What the board catches
Production AI fails in ways dashboards for ordinary software never see. Four of them, and what the record shows instead.
Silent regression
A prompt tweak ships on Friday; answer quality drops for a week before anyone notices.
boardThe eval suite fails, the release API returns fail, the deploy never happens.
Evidence scramble
An enterprise review lands with a 14-day deadline and the evidence lives in six tools.
boardSealed bundles export in one click, checksummed, with a public verify endpoint.
Unowned system
Nobody remembers who owns the reranker until the auditor asks.
boardThe registry requires an owner, risk tier, data classes, and a review cadence.
LOP · the loop
Four stations, one record
Observe, evaluate, govern, prove. Each station writes to the same ledger, so the last one is never a scramble.
01 · Observe
14.2k
traces / day
Drop the SDK in your serving layer. Trace ingest p95 budget is sub-100ms, with no sampling for high-risk paths.
02 · Evaluate
96%
release gate pass
Deterministic checks plus model-graded rubrics. Schedule daily; gate deploys on regressions.
03 · Govern
8
high-risk systems
Every AI system mapped to an owner, risk tier, data classes, and a review cadence, all auditable.
04 · Prove
31
audit artifacts ready
Immutable bundles with checksums. Send to enterprise buyers, auditors, or regulators in one click.
example data: synthetic
WIR · integration
Wired in an afternoon
Drop the Node SDK into your serving layer, or point any OTLP/HTTP exporter at the ingest endpoint. Read-scoped keys gate releases in CI. Nothing to host.
serving-layer.ts
import { AegisClient } from '@aegisrail/sdk'
const aegis = new AegisClient({
apiKey: process.env.AEGISRAIL_API_KEY!,
baseUrl: 'https://your-workspace.example',
})
await aegis.ingestTrace({
systemId: 'sys_support_agent',
provider: 'openai',
model: 'gpt-5.2',
latencyMs: 1240,
costUsd: 0.0042,
status: 'success',
riskFlags: [],
})ci-gate.ts
// CI, before deploy: read-scoped key
const gate = await aegis.getEvalGate('suite_refund_policy')
if (!gate.ok || gate.gateStatus === 'fail') {
process.exit(1) // the release never ships
}full reference: /docs/sdk
TRS · trust surfaces
The parts your buyers see
The board is for your team. These three surfaces face outward, mid-security-review.
Public trust asset
Public Trust Center
/trust/your-slug
A search-indexed page with Trust Score, systems, incident history, and frameworks; the link you paste into the review thread.
Audit integrity
Tamper-evident proof chain
verify: public
Approvals, exports, eval runs, and resolutions in an append-only SHA-256 chain anyone can check.
EU AI Act
Annex IV technical file
8 sections mapped
A versioned EU AI Act technical file generated from recorded system data, ready before high-risk obligations apply in December 2027.
Frameworks supported for evidence mapping: SOC 2 · ISO 27001 · EU AI Act (Annex IV) · NIST AI RMF · GDPR
SPC · spec sheet
Engineering targets, not testimonials
No customer metrics are quoted here because we do not invent them. These are the targets the platform is built and tested against.
100ms
trace ingest p95 budget
from SDK call to queryable trace
0
regressions past the gate
failing suites block the release API
0-100
Trust Score per system
A-D grade from live posture
1 click
evidence handoff
auditor request to sealed bundle
PLN · plans
Start internal. Publish trust when it matters.
Every plan gets the internal Trust Score and proof-chain integrity. Growth unlocks the public Trust Center and EU AI Act technical files; Scale adds signed evidence exports for enterprise reviews.
| Capability | Free$0 | Starter$19/mo | recommendedGrowth$49/mo | Scale$199/mo |
|---|---|---|---|---|
| AI systems | 1 | 3 | 10 | Unlimited |
| Traces / month | 5,000 | 25,000 | 250,000 | Unlimited |
| Internal Trust Score | ||||
| Tamper-evident proof chain | ||||
| Public Trust Center | ||||
| Annex IV technical file | ||||
| Signed evidence exports | ||||
| Start Free | Start Starter | Start Growth | Start Scale |
Founding pricing: launch rates locked for 12 months for early workspaces.
Public trust features follow server-side entitlements; a downgrade withdraws public access automatically.
LOG · asked before sign-up
Questions from the floor
Answered straight, and nothing here you cannot verify in the product.
Still have questions? support@eastbase.studio
> 01What is AegisRail?
One operations console for the teams that run AI in production. It connects trace observability, eval and regression gates, an AI system registry, incident tracking, and tamper-evident, audit-ready evidence into a single loop: observe, evaluate, govern, prove.
> 02Do I have to send you my raw prompts and model outputs?
No. You send traces from your serving layer with the Node SDK or any OTLP/HTTP exporter. Raw prompt and response storage is off by default; metadata, cost, latency, and risk flags power the product. You decide what is retained per workspace.
Read the SDK docs> 03How long does setup take?
Create a workspace, drop the SDK in or point an existing OTLP exporter at the ingest endpoint, and you can ship your first sealed evidence bundle in an afternoon. There is no agent to host and no infrastructure to run.
> 04Does AegisRail use AI, and will my data train a model?
Eval suites run on a deterministic runner by default. When you enable model-graded rubrics, they use Google Gemini or OpenAI judges. We do not sell customer data and do not use your trace content to train third-party models.
> 05What do I get on free versus paid plans?
Every plan includes the internal Trust Score and the tamper-evident proof chain. Starter adds real eval runs and compliance documents. Growth publishes the public Trust Center and EU AI Act technical files. Scale adds signed evidence exports and expiring share links for enterprise reviews.
Compare plans> 06Can I cancel anytime, and do you offer refunds?
Yes. You can cancel a paid plan whenever you like and keep access until the end of the period you already paid for. Billing runs through Lemon Squeezy, our merchant of record. Refund eligibility and the cancellation flow are covered in the Refund & Cancellation Policy.
Refund & Cancellation Policy> 07Can I export or delete my data?
Yes. Evidence bundles export as PDF and machine-readable files, and trace data exports as CSV on eligible plans. Workspace export and deletion requests are handled through support, subject to records we must retain for billing, security, or legal reasons.
Privacy Policy> 08How is my data secured, and which frameworks do you support?
Access is organization-scoped, API keys are stored hashed, outbound webhook secrets are encrypted at rest, and every change lands in an append-only audit log. Evidence is sealed into a SHA-256 proof chain with a public verification endpoint. Frameworks supported for evidence mapping: SOC 2, ISO 27001, EU AI Act (Annex IV), NIST AI RMF, and GDPR.
Call the board green.
Free plan includes the internal Trust Score and the proof chain. First trace to first sealed evidence bundle in an afternoon.