OUTAI S1 · RESEARCH
S1
System-One decision intelligence.
Compact, low-latency intelligence for routing, ranking, scoring, gating, escalation and outcome verification.
ResearchRelease status
Decision / TextPrimary modality
EvidenceRelease gate
VersionedDeployment contract
Hero product · OUTAI S1
ResearchSystem-One decision intelligence.
Deployment boundary
Local, private, sovereign and CPU-capable targets where validated.
Availability is not implied by this page. Public endpoints, pricing and limits will be published only after the relevant release is cleared.
Evaluation
Decision accuracy, ranking, calibration, abstention, latency, throughput, quantization impact and sovereign-routing behavior.
Arabic & GCC language
Modern Standard Arabic, Gulf Arabic, English-Arabic code-switching and regional enterprise terminology are explicit evaluation axes.
Sovereignty boundary
Deterministic platform controls establish the permitted runtime set first. OUTAI S1 may rank and select only within that approved set; it cannot expand or override it.
Release discipline
Research → Preview → Cleared → Retired, with model lineage and evidence attached to each cleared release.
OUTAI S1Fast decisions.
Fast decisions.
Deep reasoning only when needed.
Not every enterprise decision requires a large language model. OUTAI S1 is EADPAG’s compact System-One decision-model programme for fast, structured operational choices while larger reasoning systems remain available for complex work.
Route
Models, agents, missions, skills and tools.
Gate
Retrieval, context, confidence and evidence.
Escalate
Move uncertain or complex work to deeper reasoning or human review.
Verify
Assist with recovery decisions and outcome verification.
System-One + System-TwoSmall model first.
Small model first.
Deep reasoning on demand.
OUTAI S1 acts as the fast decision layer. Complex reasoning can then be delegated to OUTAI reasoning models, private enterprise models or approved third-party models.
Rules / Cache
↓OUTAI S1
↓Structured decision
OUTAI S1
↓Complexity or uncertainty detected
↓System-Two reasoning model
↓Governed execution
↓Verified outcome
Efficient local deploymentIntelligence without
Intelligence without
datacenter hardware.
OUTAI S1 is being designed from the beginning for compact deployment, including quantized variants for standard enterprise hardware and CPU-only environments where validated performance permits.
Nano
Compact decision intelligence for laptops, edge systems and low-resource deployments.
Core
The standard enterprise decision model for local and private deployments.
Enterprise
A higher-capability decision model for demanding workloads and greater concurrency.
GPU optional
Dedicated GPU infrastructure is not intended to be a requirement for baseline OUTAI S1 deployment.
Quantization-first engineeringEfficiency is part
Efficiency is part
of the lifecycle.
The programme evaluates INT8, INT4, GGUF, AWQ, GPTQ and hardware-native formats where appropriate. Every compressed release remains subject to quality, calibration and reproducibility gates.
Measured compression
A smaller model is released only when measured decision quality remains within its approved operating boundary.
Reproducible evaluation
Compression profiles are tested alongside structured-output reliability, latency, throughput and resource use.
Deployment boundary
Each release records the hardware and operating conditions under which it is approved.
Provider-neutral intelligenceLocal decision intelligence.
Local decision intelligence.
Freedom at the reasoning layer.
OUTAI S1 is not designed to lock enterprises into a single model provider. Provider neutrality does not override sovereignty. OUTAI S1 operates only across models, runtimes, agents and tools permitted by the governing Mission contract. Use the smallest capable and permitted intelligence for each decision, within the required sovereignty and governance boundary, and escalate only when deeper reasoning is actually required.
Eligible reasoning backends only when permitted by the governing Mission contract.
OUTAI reasoning modelsPrivate / sovereign modelsOpenAIAnthropic ClaudeGoogle GeminiOther approved enterprise models
Sovereign decision intelligenceOptimize within the boundary.
Optimize within the boundary.
Never around it.
Deterministic sovereignty controls first establish the eligible models, runtimes, agents, tools, storage and connectors according to jurisdiction, residency, security and administrative-control requirements. OUTAI S1 then performs fast ranking and selection only within that approved candidate set.
Permitted-set optimization
Rank permitted models, agents, tools and runtimes without expanding the authorized execution boundary.
Local context minimization
Select the minimum relevant context before deeper reasoning to reduce unnecessary data movement and exposure.
Compliant fallback
Choose another permitted runtime when available rather than silently crossing a sovereignty boundary.
Abstain or escalate
When no compliant runtime is available, return a stop or escalation decision instead of routing to a non-compliant provider or region.
Advisory anomaly detection
Surface potential sovereignty anomalies for deterministic validation and enforcement. S1 does not independently authorize HOLD, HALT or isolation actions.
Canonical rule: OUTAI S1 may optimize within sovereignty; it may never optimize around sovereignty. It may rank the permitted set, but it may never expand the permitted set.
Enterprise decision intelligenceA specialist for
A specialist for
operational decisions.
OUTAI S1 is being developed for high-frequency decisions where compact, structured intelligence can reduce latency, cost and unnecessary large-model calls.
Model routing
Select the appropriate permitted model based on capability, quality, latency, privacy, deployment constraints and cost.
Agent routing
Select the appropriate AI worker, specialist or automation path from the eligible candidate set.
Mission routing
Match objectives to known executable missions and workflows.
Skill routing
Identify and rank the capabilities required to complete work.
Tool routing
Select the appropriate API, enterprise connector, browser, computer-control mechanism or workflow.
Retrieval gating
Determine whether external knowledge or enterprise retrieval is actually required.
Context selection
Identify the minimum relevant context required for the next operation.
Escalation
Recognize when a decision exceeds System-One capability and requires deeper reasoning or human review.
Recovery
Support retry, alternative-path and compensation decisions after execution failures.
Evidence assessment
Evaluate whether available evidence is sufficient to support a claimed outcome.
Outcome verification
Assist in determining whether executed work actually achieved the intended objective.
Confidence, uncertainty and abstentionA decision model must
A decision model must
know when not to decide.
OUTAI S1 is being designed around calibrated confidence rather than forced prediction. Uncertainty is treated as useful information; when confidence is insufficient, the correct result may be escalation rather than prediction.
PROCEEDRETRIEVEESCALATEASK APPROVALINSUFFICIENT CONTEXTINSUFFICIENT EVIDENCELOW CONFIDENCEOUT OF DISTRIBUTIONREQUIRES HUMANNO COMPLIANT RUNTIMECROSS-BORDER AUTH REQUIRED
Intelligence is not authorization.
Governed executionRecommend fast.
Recommend fast.
Authorize deterministically.
OUTAI S1 can recommend actions. It does not automatically authorize them. Sovereignty, jurisdiction, residency, identity, entitlement, security policy and execution controls remain deterministic authorities. OUTAI S1 may rank or select only from eligible resources supplied by those authorities, and model confidence can never override them.
Privacy-aware context
Local context selection can reduce unnecessary data exposure, token consumption, inference cost, latency and irrelevant context.
Sovereign deployment
Decision intelligence can remain local while approved reasoning systems are selected according to jurisdiction, residency, administrative-control and enterprise deployment policy.
Explicit authority
Security, identity, permissions and policy stay outside model confidence and remain deterministic.
Learning from verified outcomesBetter decisions require
Better decisions require
more than imitation.
OUTAI S1 is intended to learn from synthetic examples and teacher models, and eventually from validated operational outcomes under explicit training eligibility, provenance, customer boundaries and data governance.
Decision→Action→Evidence→Actual result→Verified outcome→Evaluation→Improved future model
Evidence before claimsBenchmark first.
Benchmark first.
Claim second.
OUTAI S1 is being evaluated against deterministic routing, embedding-based selection, generic small-model prompting, specialist decision models, fine-tuned compact language models and large reasoning models.
Decision quality
Accuracy, ranking quality, calibration, abstention and structured-output reliability.
Efficiency
Latency, throughput, RAM / VRAM requirements and inference cost.
Compression impact
Measured quality changes across approved quantized variants.
System impact
External model calls avoided without violating the approved operating boundary.
No invented benchmark scores.
No unsupported superiority claims.
Measurements first.
Built for ServAI. Designed beyond it.Fast decision layer.
Fast decision layer.
Reusable intelligence.
ServAI is the first major enterprise platform intended to consume OUTAI S1 beneath missions, skills, agents, tools, retrieval and model routing, while ServAI continues to provide governance, execution and verified outcomes.
OUTAI S1 is the public OUTAI model programme today and is designed to be reusable across EADPAG products and controlled enterprise deployments. Further OUTAI specialists will be listed publicly only after they clear their own release gates.
Built to decide,
trained to verify,
released only with evidence.