A critical assessment of the $24M Series A led by Norwest (Assaf Harel) with Snowflake Ventures — building the "context graph for enterprise AI" that arms agents with business context. Strong repeat founders, real customers, and a structural problem: every platform it partners with is building, buying, or open-sourcing the same layer.
The existential question. Snowflake ships Semantic Views and Cortex Analyst natively, acquired Select Star, and funded AtScale. The dbt/Fivetran merger ($600M combined ARR) open-sourced "Agents Schema" — a free, standards-based attack on exactly this layer. Nvidia bought competitor Illumex for just $60–75M.
The lead investor's own case study complicates the claim. Norwest reports The Weather Company went from 40–45% accuracy "out of the box" to 85%+ only "after a focused refinement process." That's services-assisted onboarding — and 85% still means roughly 1 in 7 answers wrong.
The team. Henkin has a real exit (Kontera → Singtel, 2014) and 15+ years building data/AI products with his co-founders. Repeat-founder quality is the strongest pillar of this round — stronger than the category position.
Key Finding: Jedify is a well-built company in a layer the platforms are actively absorbing. Its closest partner — Snowflake — is simultaneously its investor, its distribution channel, and its competitor, having already acquired one rival and funded another. The Snowflake relationship is both Jedify's biggest asset and its biggest tell.
The category's recent history reads as a countdown. Every quarter, another independent context player gets built natively, bought cheaply, or open-sourced away.
Snowflake Ventures leads AtScale's round — a second, competing semantic-layer bet by Jedify's own strategic investor.
Nvidia acquires Illumex — a direct "semantic fabric" competitor — for $60–75M, at or below Jedify's likely post-money.
Snowflake acquires Select Star (metadata context) for its Horizon Catalog and ships Semantic Views + Cortex Analyst natively.
Fivetran and dbt Labs complete their merger (~$600M combined ARR) and launch "Agents Schema" — an open-source standard for agentic context.
Jedify raises $24M to be the independent context layer — nine days after the open standard launched.
Jedify's product requires read access to warehouses, CRMs, Slack, code bases, and meeting recordings — a single aggregation point for a company's most sensitive context. Yet the public privacy policy covers only the marketing website, no public subprocessor list discloses which LLM providers process customer data, and the trust center requires sign-in. For a "model-agnostic" company, there is no public statement of where inference actually happens.
The repositioning tell: Snowflake's own June 2025 Startup Spotlight described Jedify as conversational BI — "Ask Jedify," self-service analytics. Twelve months later it's "the context graph for enterprise AI." Same graph, new narrative — a rational move that also reveals how fast this category's framing is being repriced by the agent wave.
The product's headline numbers and the lead investor's case study tell two slightly different stories — and the difference is the business model.
"40% accuracy improvement." "70% fewer tokens with same accuracy." An autonomous context graph that "gets smarter with every interaction." No methodology, benchmark set, or baseline has been published for any of these figures.
Norwest: The Weather Company went from 40–45% out-of-the-box accuracy to 85%+ — "after a focused refinement process." That is services-assisted onboarding, and it undercuts the core differentiation against manually maintained semantic layers.
The graph is built by "mining query logs at scale" (Norwest). Jedify's target mid-market — gaming, industrials, CPG — is precisely where warehouse query logs are thinnest. The autonomy story works best at companies that need it least.
And 85% is not a finish line. In production analytics, 85% accuracy means roughly one in seven answers is wrong — tolerable for exploration, untenable for the autonomous decision-driving agents the pitch describes. The gap between "demo accuracy" and "decision accuracy" is where this category's churn lives.
Seven structural risks the $24M does not resolve.
Snowflake, Databricks, OpenAI, and Anthropic are all building context capability natively; Snowflake — Jedify's closest partner — has already acquired one competitor and funded another. Context may be a feature, not a product.
dbt/Fivetran's free Agents Schema and MCP itself push context definition toward open, vendor-neutral standards — collapsing the pricing power of a proprietary graph.
Marketing says autonomous; the flagship case study required "focused refinement" to reach 85% — a services-heavy motion that scales poorly and undercuts the differentiation.
One vendor aggregating warehouse + CRM + Slack + meeting-recording access, with no public LLM subprocessor disclosure and a gated trust center. A single failed enterprise security review is existential at this stage.
Glean ($7.2B valuation, ~$200M ARR), dbt/Fivetran ($600M ARR), WisdomAI ($73M, Nvidia-backed) all converge on the same buyer with more capital and distribution.
No ARR disclosed in ~2.5 years; 10–20 customers weighted toward Israeli founder-network companies; repositioned from conversational BI to agent infrastructure within 12 months.
The category's only realized exits — Illumex to Nvidia ($60–75M) and Select Star to Snowflake (undisclosed tuck-in) — suggest acquirers price this layer as a feature, roughly at or below Jedify's likely current post-money.
Jedify is the strongest team in this week's funding cohort attacking the weakest structural position. The context-graph thesis is real and the agent wave makes it urgent — which is exactly why Snowflake builds it, Nvidia bought it, and dbt/Fivetran open-sourced it. The $24M buys three years to prove that independent context infrastructure outruns platform absorption. Watch the Snowflake relationship: the day it ships a native context graph, Jedify's biggest channel becomes its biggest competitor.
Based entirely on publicly available information, including the TechCrunch announcement of June 10, 2026. Figures labeled est. are derived, not disclosed.
