Service Level Agreement (SLA)

    Service Level Agreement (SLA)  – Available for Enterprise Version of OdysseyAI Only (Not Available for Versational)

    This SLA, aspart of the Terms of Service, outlines InteliGems Labs ("Provider") productperformance commitments for Odyssey AI (the "Service") to you("Customer").

    I. Definitions


    1. Accuracy

    Factual Consistency Score (FCS) – Percentage of responses matching verified ground truth data, measured against MMLU benchmark subset. MMLU = v2024.1 STEM subset (Mathematics/Physics/Chemistry).

    2. Inference Latency

    Time elapsed from request submission to the first token of the response, excluding network latency outside the Provider's control. It is typically measured by recording the time elapsed from the moment an input is fed into the model to the moment the output is produced. Latency will be measured under "Normal Operating Conditions."

    3. Uptime

    Percentage of time the Service is operational and available per calendar month, excluding Scheduled Downtime and Exclusions.

    4. Monthly recurring charges (MRC)

    Calculations are based on subscriptions  for the affected Service.


    5. Normal Operating Conditions

    Refers to the typical usage patterns and load experienced by the Service, excluding periods of peak demand, scheduled maintenance, or force majeure events. Example:

    1. (a) AWS EC2 c6i.8xlarge instances | 64GB RAM | PyTorch 2.2.1
    2. (b) Load tested at 1,000 RPS
    3. (c) ≤500requests/sec | ≤5MB payloads | ≤1,000 concurrent users

    Service Commitments



    1. Accuracy:

    1. (a) 98% FCS on 1,000 validated MMLU queries per domain
    2. (b) For domain-specific applications, accuracy rates may vary and will be specified in separate        agreements or addenda. HELM benchmarks used for domain SLAs
    3. (c) FCS measured against MMLU v2024.1 benchmark(1,000 queries per domain)
    4. (d) FCS = (Expert-Validated Correct Responses ÷Total Responses) × 100
    5. (e) Percentage of responses matching expert-validated ground truth from MMLU v2024.1 STEM        subset(Mathematics/Physics/Chemistry) using BERTScore ≥0.85
    6. (f) Expert Validated = 3 Masters -level experts validate each response, majority vote resolves disputes

    Threshold:

    FCS below 98% for 3 consecutive days across ≥1,000 queries


    Remedies:

    • 98-97% Accuracy = 5% of MRC Service Credit
    • 96.9-95% Accuracy = 10% of MRC Service Credit
    • <95% Accuracy = 15% of MRC Service credit
    • *For accuracy rates below 90%, customers may request additional support

    Conditions:

    1. (a) Applies only to tasks with statistically significant sample sizes (e.g., ≥1,000 queries/month).
    2. (b) BERTScore v1.0.3 (roberta-base)
    3. (c) Excludes: 1) Unsupported languages/regions 2) Hallucinations in creative mode 3) Responses        beyond documented system knowledge cutoff
    4. (d) Customers receive monthly FCS reports withconfusion matrices
    5. Example
    6. 3 days of 97.2% FCS across 1,200 queries → 5% credit

    2. Inference Latency (non-Agentic)

    • (a) P90 Latency: 90% of requests will have an inference latency of less than 300 milliseconds.
    • (b) P50 Latency: 50% of requests will have an inference latency of less than 150 milliseconds.
    • (c) These latency metrics apply under the defined "Normal Operating Conditions."
    • (d) AWS X-Ray with trace sampling (10% requests)

    Threshold:

    P90 latency exceeds 300ms in >10%of requests during any 24-hour period.

    Remedies:

    • 1-4 hours Excess Latency Duration = 2% of MRC Service Credit
    • 4-8 hours Excess Latency Duration = 5% of MRC Service Credit
    • >8 hours Excess Latency Duration = 10% of MRC Service Credit

    Exclusions:

    1. (a) Excludes latency caused by customer network issues, large payloads (>10MB), or third-party        integrations
    2. (b) Excludes Customer network latency >10ms
    3. (c) Excludes Third-party integrations (e.g., Salesforce/MS Dynamics)
    4. (d) Excludes credits for downtime/errors duringscheduled maintenance, force majeure, or security        incidents
    5. (e) Credits apply per incident, with a monthly cap of 15% of MRC
    6. Example Calculation:
    7. (a) A customer paying $10,000/month experiences12 hours of P90 latency breaches (10% credit = $1,000) and 3 days of 97.2% FCS across 1,200 queries (10% credit = $1,000).
    8. (b) Total credits = $2,000 (20% of MRC), but capped at 15% ($1,500) due to monthly limit

    3. Uptime:

    • The Service will maintain a monthly uptime percentage of at least 99.9%, excluding scheduled maintenance or downtime caused by factors outside InteliGems Labs's control.
    • (a) Uptime Calculation:
               TotalMinutesinMonth−DowntimeMinutes)/TotalMinutesinMonth×100(Total Minutes inMonth -          Downtime Minutes) / Total Minutes in Month \times 100(Total          MinutesinMonth−DowntimeMinutes)/TotalMinutesinMonth×100
    • (b) Health Checks: HTTP 200 to /healthcheck every 30s | 5 consecutive failures = downtime
    • (c) Uptime % = (Total minutes - Downtime minutes) / Total minutes × 100

    Remedies:

    • 99.0%-99.9% Monthly uptime percentage = 10% service credit
    • 95.0%-98.9% Monthly uptime percentage = 25% service credit
    • Below 95% Monthly uptime percentage = 100% service credit
    • For accuracy rates below 90%, customers may request additional support

    II. Measurement, Monitoring and Reporting

    1. InteliGems Labs will monitor performance metrics in real-time and provide periodic     reports to customers upon request.

    2. Customers may independently verify accuracy rates and latency using provided APIs or monitoring tools.

    III. Scheduled Maintenance

    1. Provider will notify Customers at least 48hours in advance via email and website posting for any maintenance impacting Service availability. A reminder notification will be sent 24 hours before the scheduled maintenance.

    2. Customers are responsible for checking their registered email and the InteliGems Labs website for maintenance notification

    3. Notices are posted on this website (https://www.inteligems.io/trust-center#security-and-compliance) at least 48 hours-in-advance.

    IV. Exclusions

    This SLA does not cover service disruptions caused by:

    • 1. Customer-side network issues or third-party failures
    • 2. Force majeure (acts of God, war, terrorism, and other uncontrollable events).
    • 3. Customer misuse, such as exceeding usage limits or sending malformed requests.
    • 4. Third-party services not provided by Provider.
    • 5. Beta features.
    • 6. Issues arising from Customer’s data.
    • 7. Modifications made by the Customer.
    • 8. Underlying cloud infrastructure failures (AWS/Google Cloud) beyond Provider’s control.
    • 9. Unsupported data formats or languages.
    • 10. Customer-side infrastructure failures.
    • 11. Prompt injections > 10k tokens (flagged via regex patterns)
    • 12. Attacks on the Service (e.g., DDoS) mitigated within 1 hour.
    • 13. Scheduled maintenance (e.g., InteliGems Labs pre-announced updates)
    • 14. Data incidents governed by SOC 2 controls and at https://www.inteligems.io/trust-center#privacy-policy
    • 15. Data Protection section with encryption, retention, and breach notification refer at https://www.inteligems.io/trust-center#privacy-policy

    V. Credits

    1. Credits are not automatic; customers must submit claims via InteliGems Labs Support with outage details and logs

    2. Credits apply only to the affected service’s monthly fees, excluding taxes or one-time charges

    3. Credits capped at25% of total MRC per month

    4. Total credits capped at 50% of monthly fees or 6x MRC

    VI. Support

    1. Business Hours Available:

    • 8.30 am BST – 4.30 pm BST (UK Support)
    • 6am PDT – 3 pm PDT (US Support

    2. Incident Reporting:

    3.  Response Times:

    • Provider aims to respond within 4 hours for P1 severity incidents during business hours.

    VII. Amendments

    Provider may modify this SLA by providing written notice to Customer at least 30 days prior to the effective date of the modification. Continued use of the Service after the effective date constitutes acceptance of the modified SLA.

    VII. Governing Law

    This SLA is governed by the laws of Nevada, USA. Disputes resolved through binding arbitration in Reno, NV per AAA rule. Arbitration administered by AAA under Consumer Rule.

    IX. Acceptance

    By using the Service, the Customer acknowledges and accepts the terms of this SLA.