{
  "schema_version": "2.0",
  "slug": "kubit",
  "editor": "Hlido Editor",
  "editorial_method": "public-surface-tier-1+editorial-narrative-v2",
  "methodology_version": "2026.05",
  "methodology_url": "/methodology/public-surface-tier-1/",
  "engine": "public-surface",
  "evidence_tier": "screenshot",
  "source": "r-publish-editorial-enrich",
  "run_id": "run-kubit-v2-2026-08-18",
  "run_at": "2026-08-18T08:45:00Z",
  "staleness_after": "2026-11-16",
  "next_review_due_at": "2026-11-16",
  "signature_pending": true,
  "name": "Kubit",
  "agent_url": "https://kubit.ai",
  "category": "Frameworks & Eval",
  "score": 74,
  "laddoo_score": 74,
  "tier": "STEADY",
  "confidence": "medium",
  "hlido_opinion": {
    "headline": "A mature warehouse-native product-analytics company extending into agent observability — its real edge is joining agent traces to user behaviour in your own warehouse, not another dashboard that stops at the LLM.",
    "body": "Kubit's pitch lands on a gap most agent-observability tools leave open: infrastructure tracing tells you whether the LLM was fast, product analytics tells you whether the user clicked, and neither explains why an AI feature actually failed a user. Kubit connects the two — enriching agent traces (prompts, tool calls, model output) with intent and sentiment and correlating them with real user outcomes like re-prompts, drop-off, conversion and retention. The differentiator that reads as substantive rather than slogan is the architecture: warehouse-native and 'bring your own warehouse', so PII never leaves your store; ingestion via open standards (OpenTelemetry for traces, CDP or direct warehouse query for events); and no proprietary black box. For the agent thesis specifically, two things matter — it is 'headless for coding agents' (feed Claude Code the exact behavioural insight to debug a UX issue) and it embeds product analytics into agent loops as a verifier via MCP, which is a genuinely useful pattern beyond unit tests. Trust signals are strong for enterprise buyers: named customers (Miro, GameChanger, Serko, Vix) with attributed testimonials and a Trust Center. The caveats are those of a company adding an agent story onto an established analytics core: the agent-specific capabilities are newer than the mature product-analytics base, the 'billions of rows' and correlation claims are not independently benchmarked on the surface, and pricing is not public. But the warehouse-native, open-standards posture is a real position in a category full of lock-in, and the trace-to-behaviour join is a legitimately differentiated answer.",
    "voice": "Hlido Editor",
    "as_of": "2026-08-18",
    "editor_signature_pending": true
  },
  "tier_rationale": "STEADY (74) because Kubit brings a mature, warehouse-native analytics platform with real enterprise customers and an open-standards (OTel/CDP/SQL) architecture to a genuine gap — correlating agent traces with user outcomes — plus concrete agent hooks (headless insight for coding agents, MCP as an in-loop verifier). Not VITAL because the agent-specific layer is newer than the underlying product-analytics core, the scale and correlation claims are not independently benchmarked on the surface, and pricing and agent-analytics adoption specifics are not public.",
  "what_it_does_well": [
    "Joins agent traces to real user behaviour (re-prompts, drop-off, conversion, retention) instead of stopping at the LLM",
    "Runs warehouse-native / bring-your-own-warehouse so PII and sensitive data never leave your store",
    "Ingests via open standards — OpenTelemetry, CDP, direct SQL — with no proprietary black box or vendor lock-in",
    "Exposes analytics to coding agents headlessly and as an in-loop verifier via MCP, beyond unit tests",
    "Carries strong enterprise trust signals: named customers with attributed testimonials and a Trust Center"
  ],
  "what_it_fails_at": [
    "The agent-analytics capability is newer than, and layered onto, an established product-analytics core",
    "Scale and correlation claims ('billions of rows') are not independently benchmarked on the captured surface",
    "No public pricing, so cost and commercial fit cannot be assessed from the surface",
    "Requires a data warehouse and OTel/CDP plumbing — meaningful setup, not a drop-in",
    "Agent-specific adoption evidence is thinner than the general product-analytics testimonials"
  ],
  "best_for": [
    "Product and data teams shipping AI features who need to know why users re-prompt or churn, not just that they did",
    "Enterprises with a data warehouse and OTel/CDP pipeline wanting agent + user analytics without moving PII",
    "Coding-agent workflows that want behavioural insight and analytics-as-verifier fed in via MCP",
    "Organisations avoiding proprietary-black-box analytics in favour of open standards"
  ],
  "not_recommended_for": [
    "Small teams without a warehouse or the appetite to wire OTel/CDP ingestion",
    "Buyers who require public pricing before evaluating",
    "Anyone wanting a zero-setup, drop-in dashboard rather than a warehouse-native platform",
    "Pure LLM-latency monitoring needs where user-behaviour correlation is irrelevant"
  ],
  "red_flags": [],
  "compared_to": [
    {
      "slug": "langsmith",
      "verdict_diff": "LangSmith and Langfuse optimise agent traces largely in isolation from downstream user behaviour; Kubit's whole thesis is joining those traces to user outcomes in your own warehouse. Kubit wins when the question is 'did this agent behaviour actually help or hurt the user'; the trace-native tools win for deep prompt/chain debugging and eval tooling closer to the model.",
      "preferred_for_axis": "trace-to-user-outcome-correlation"
    },
    {
      "slug": "langfuse",
      "verdict_diff": "Langfuse is open-source, trace- and eval-centric observability for LLM apps; Kubit is warehouse-native product analytics that ingests agent traces (via OTel) and correlates them with user behaviour and business dimensions. Use Langfuse for model-side tracing and evals; use Kubit when the analysis must span agent actions and real user journeys without moving PII out of your warehouse.",
      "preferred_for_axis": "warehouse-native-product-analytics"
    }
  ],
  "evidence_urls": [
    {
      "claim": "Correlates agent traces with user behaviour and outcomes (re-prompts, retention)",
      "source": "https://kubit.ai/",
      "tested_at": "2026-08-18",
      "verified": true
    },
    {
      "claim": "Warehouse-native / bring-your-own-warehouse; PII stays in the warehouse",
      "source": "https://kubit.ai/",
      "tested_at": "2026-08-18",
      "verified": true
    },
    {
      "claim": "Open standards ingestion: OTel, CDP, SQL; MCP as in-loop verifier",
      "source": "https://kubit.ai/",
      "tested_at": "2026-08-18",
      "verified": true
    },
    {
      "claim": "Named enterprise customers (Miro, GameChanger, Serko, Vix)",
      "source": "https://kubit.ai/",
      "tested_at": "2026-08-18",
      "verified": true
    },
    {
      "claim": "Independently benchmarked scale / correlation performance",
      "source": "https://kubit.ai/",
      "tested_at": "2026-08-18",
      "verified": false
    }
  ],
  "agent_relevance": {
    "has_api": true,
    "has_cli": false,
    "has_mcp": true,
    "has_webhook": false,
    "has_sdk": true,
    "behavioral_testable": false,
    "agent_integration_path": "Two agent-facing paths on the surface: it ingests agent traces via OpenTelemetry, and it exposes product analytics to agents through MCP so a coding agent can pull behavioural insight to debug a feature or use analytics as an in-loop verifier. It is analytics infrastructure agents feed into and query, not an agent itself; setup requires a warehouse plus OTel/CDP wiring.",
    "agent_friendly_score": 7
  },
  "marking_signal": {
    "not_applicable": true,
    "reason": "Kubit is an analytics/observability platform for agents and users and does not generate synthetic content, so Article-50 output-marking obligations do not attach.",
    "checked_at": "2026-08-18"
  },
  "summary": "A mature warehouse-native product-analytics company extending into agent observability — its real edge is joining agent traces to user behaviour in your own warehouse, not another dashboard that stops at the LLM.",
  "_summary_deprecation_note": "Field kept as a v1-compatibility alias of hlido_opinion.headline. New consumers should read hlido_opinion.{headline,body,voice,as_of}."
}