News · Announcements

Introducing Koba AI

Koba logo
Ink illustration of a butterfly

Motivation and problem formulation

Professional sports desks already have dashboards. What they lack is an assistant that can (a) answer quickly between meetings, (b) run longer research when a narrative or deal moves, and (c) stay sponsorship- and brand-aware without collapsing into generic web chat. We formulate Koba as a mid-tier sports intelligence agent: tool-using, domain-specialized, and oriented to everyday professional workflows rather than only flagship scout-depth reasoning (that flagship stats tier is StatGost).

Koba agent tool-use loop (architecture schematic)User goalPlanToolsRetrieveSynthesizeCite
Figure 1. Architecture schematic (qualitative) of Koba’s agent loop from user goal through cited synthesis.

Architecture schematic (qualitative).

System architecture

Koba is positioned as “our most agentic mid-tier model yet, with top-tier intelligence for sports and everyday professional workflows.” Capability bullets on announcement: Quick Search; Deep Research; Sponsor Analytics; Brand Advisory; Live Dashboard. Koba is only available in Eagle Eye dashboards and products (for example Sponsorlytix and Dantani Sports surfaces where the agent is embedded). It is not offered as a standalone consumer app, and there is no separate public Koba product site for sign-up. Access is through the Eagle Eye suite products that include Koba.

Architecturally we treat Koba as an LLM agent with sports-domain tooling: search over sports/commercial corpora, deeper multi-step research, sponsorship-aware analytics calls, brand-advisory prompts grounded in measurement vocabulary, and a dashboard working surface for ongoing threads.

Workflow × tool emphasis (illustrative)RetrieveMulti-hopNumbersAdvicePersistQuick SearchDeep ResearchSponsor AnalyticsBrand AdvisoryLive Dash
Figure 2. Illustrative evaluation profile — not a published benchmark. Intended emphasis of tool classes by Koba workflow mode.

Illustrative evaluation profile — not a published benchmark.

Data and methods

Koba reasons over sports, esports, sponsorship, and gaming professional topics. Grounding should prefer suite measurement and monitoring artifacts when the user has access, and clearly mark when answers are general knowledge vs retrieved evidence. We do not publish a model card with parameter counts, training mix, or public leaderboard scores in this article.

30 Jun 2026Koba announcement date
Mid-tierPositioning vs flagship StatGost
5Named capability pillars on launch card
AgenticTool-using professional workflows
Illustrative agent quality axes (relative emphasis)SpeedDepthGroundingSponsorSafetyUX
Figure 3. Illustrative evaluation profile — not a published benchmark. Relative design emphasis across agent quality axes for mid-tier Koba.

Illustrative evaluation profile — not a published benchmark.

Comparison vs suite siblings

DimensionKobaStatGostDantani dashboard
Primary interfaceAgentic chat / workflowsStats-first flagship reasoningMonitoring UI
StrengthSearch → research → advisoryPlayer/league/finance/sponsor statsSoV / AVE / partner packs
Tier languageMid-tier agenticFlagshipPlatform product
AvailabilityOnly inside Eagle Eye dashboards & productsEagle Eye suite positioningLive Dantani / suite surfaces

Evaluation table

MetricDefinitionUnitProtocolResult
Task success (desk tasks)Human-rated completion of sports desk tasks% passInternal rubric, blinded ratersInternal eval — contact sales
Citation faithfulnessClaims supported by retrieved snippets%Claim–evidence auditInternal eval — contact sales
Tool routing accuracyCorrect tool selected for intent%Intent suite—
Time-to-useful-answerLatency to first actionable paragraphsOperational sampling—
Sponsorship term precisionCorrect use of inventory / rights terms%Domain glossary checksInternal eval — contact sales

Limitations and failure modes

  • Hallucinated transfer or injury facts when tools are skipped or corpora are stale.
  • Over-confident brand advice without measurement binds.
  • Not a standalone product — Koba only runs inside Eagle Eye dashboards and products; there is no separate Koba login or public demo site.
  • Mid-tier ≠ weak: positioning is about workflow breadth and cost/latency tradeoffs vs flagship StatGost, not a claim of inferior sports knowledge in every subtask.

Safety and responsible use

Agentic sports assistants can spread rumour, medical speculation, or market-moving claims. We expect refusal / caution patterns around minors, unverified medical data, and non-public deal terms. Tool outputs that look like measurement should be labeled as such; narrative advice should not be presented as Sponsorlytix-certified brand value.

What’s next / availability

Koba is available only inside Eagle Eye dashboards and products. There is no standalone Koba app and no plan announced here for a separate public Koba site. Use Koba where it is embedded in suite products (Sponsorlytix, Dantani Sports, and related Eagle Eye surfaces), or contact sales for product access that includes Koba.

Appendix — glossary

Agentic. Multi-step tool use toward a user goal, not single-shot completion only.

Deep Research. Longer-horizon investigation across sports and commercial topics.

Brand Advisory. Guidance for positioning and partnership questions, ideally grounded in measurement vocabulary.

Explore Koba →

← All announcements