ABS Core
Legal & Company

Service Level Agreement

Formal SLA commitments for ABS Core — uptime, response times, MTTR, RTO/RPO, and support tiers.

Service Level Agreement

Effective Date: 2026-02-01 · Version: 2.0 · Applies to: ABS Core v10.1.x+

This document defines the service-level commitments for ABS Core. It is binding for Enterprise and Sovereign licensees with an active commercial agreement. Community Edition users receive best-effort service without SLA guarantees.


Uptime Commitment

TierMonthly UptimeMax Downtime/MonthMax Downtime/Year
Enterprise (SaaS)99.95%21.9 minutes4.4 hours
Sovereign (VPC/On-Prem)Customer-managedCustomer-managedCustomer-managed
CommunityBest effortNo commitmentNo commitment

Uptime is measured as availability of POST /v1/authorize returning a decision within 300ms, excluding:

  • Scheduled maintenance windows (announced 72h in advance)
  • Force majeure events
  • Customer-caused outages (misconfigured policies, invalid API keys)

Latency SLO

PercentileCommitmentMeasured at
P50< 200msOperational load (≤ 1,000 req/s)
P99< 300msOperational load (≤ 1,000 req/s)
P99.9< 500msOperational load (≤ 1,000 req/s)

Latency SLOs apply at or below 1,000 req/s. Above this threshold, the rate limiter engages and the latency SLO no longer applies — the commitment becomes zero 5xx errors and complete audit integrity.


Support Tiers

TierChannelsSev-1 ResponseSev-2 ResponseSev-3 Response
CommunityGitHub DiscussionsBest effortBest effortBest effort
StarterEmail< 24h< 48h< 5 business days
EnterpriseDedicated Slack + Email + Portal< 1h< 4h< 24h
SovereignDedicated Slack + On-site option< 1h< 4h< 24h

Severity Definitions

LevelDefinitionExample
Sev-1Production down — governance engine returning 5xx or not responding/v1/authorize returning 500 for all requests
Sev-2Major functionality impaired, workaround existsCHI semantic analysis degraded; shadow mode active
Sev-3Non-critical issue or questionPolicy authoring question, dashboard cosmetic issue

MTTR and Recovery Commitments

ScenarioMTTR TargetMechanism
Edge node failure< 30 secondsCloudflare automatic rerouting to adjacent PoP
CHI engine timeout< 200msCircuit breaker → Fail-Safe ALLOW
Full region outage< 15 minutesCloudflare multi-region failover
Scheduled maintenance0 downtimeRolling deploys with zero-downtime strategy

RTO / RPO

ParameterCommitmentNotes
RTO< 15 minutesTime to restore full service after worst-case outage
RPO (decisions)0Decisions written to D1 before response is returned — no decision is ever lost
RPO (L2 anchoring)< 1 block cycle (~2s)L2 queue persisted; anchor completes after recovery

Data and Audit Guarantees

CommitmentDetail
Audit completeness100% of requests — ALLOW, DENY, and rate-limited — are recorded
Tamper evidenceEvery audit record is SHA-256 hash-chained to the previous
ImmutabilityAudit records are append-only; no delete or update operations exist
L2 anchoringBatches anchored to Polygon L2 within 1 block cycle (~2s) under normal conditions — Enterprise tier only
RetentionAudit records retained for minimum 7 years (configurable per compliance requirement)

SLA Credits

For Enterprise/Sovereign licensees, downtime beyond the committed uptime threshold entitles the licensee to service credits:

Monthly UptimeCredit
99.90% – 99.94%10% of monthly fee
99.00% – 99.89%25% of monthly fee
Below 99.00%50% of monthly fee

Credits are applied to the next billing cycle and do not exceed the monthly fee. Credits are the sole remedy for SLA breaches under this agreement.


Exclusions

This SLA does not apply to:

  • Downtime caused by customer-side infrastructure (network, firewall, DNS)
  • Downtime caused by upstream LLM providers (OpenAI, Anthropic, Google) when using the Magic Proxy
  • Cloudflare platform-wide incidents affecting all customers
  • Events of force majeure
  • Sovereign/On-Prem deployments managed by the customer

Verification

SLA compliance can be independently verified via:


Contact for SLA Disputes

Enterprise licensees may raise SLA disputes by emailing [email protected] with subject SLA Dispute — [workspace-id]. Disputes must be raised within 30 days of the incident.

On this page