> ## Content Index
> Fetch the complete content index at: https://nexi.fund/llms.txt
> Use this file to discover other available public pages before exploring further.

# Zeroset Raises $5.2M to Give AI Agents Real Workflow State
- URL: https://nexi.fund/zeroset-nebula-enterprise-agent-state-2026/
- Published: 2026-10-06T20:50:13.000Z
- Updated: 2026-10-06T20:50:13.000Z
- Description: Gradient-backed pre-seed for Nebula, a hierarchical state layer that lets long-horizon agents query current enterprise workflows instead of rebuilding context every turn.
- Author: Nexi.fund Labs
- Tags: AI & Infrastructure, #mode-5, #hook-number, #track-E

$5.2 million is a small number in AI infrastructure. The retrieval delta is not. On LongMemEval, Nebula returned 20% higher accuracy than Mem0, Supermemory, and naive RAG, at a median under 50 milliseconds and with fewer tokens per query.

That is the signal Zeroset is selling: not another memory store, but a state layer that lets long-horizon agents query the current workflow instead of rebuilding context from scratch every turn.

## From dropouts to closed research preview

Akshat Kannan left Stanford at 18\. William Zhang left the University of Texas at 21\. They founded Zeroset in January 2026 with a narrow claim: frontier models already handle email drafts and flight searches. They stall when the task requires knowing how a specific company actually routes approvals, exceptions, and handoffs.

On 6 October 2026 the company closed a $5.2 million pre-seed co-led by Gradient Ventures (Google’s early-stage AI fund) and 2048 Ventures, with Leblon Capital participating. Headcount sits at five. Access remains closed research preview; no broad launch date has been set.

## What Nebula actually stores

Nebula connects to the systems a workflow already uses—Microsoft 365, SharePoint, Outlook, Teams, GitHub, and workplace messaging. It does not treat every document as an isolated vector. It builds a hierarchical vector graph with three layers: semantic facts, episodic event clusters, and procedural habits.

Agents query the current state of a workflow rather than reconstructing history. On the BPI Challenge process-mining datasets the system beat all recorded baselines on next-activity prediction by more than 10%. That is a harder test than pure retrieval: it asks whether the model can anticipate the next legitimate step.

The company frames the failure mode clearly. A stale fact or contradictory instruction does not cause a single error. It becomes the starting state for the next turn, compounding rework, extra tool calls, and human intervention. Pure memory systems retrieve similar chunks. Nebula returns typed, queryable state with lineage—what is true, when it became true, and why the system believes it.

## Competitive frame

Mem0 and Zep occupy the adjacent memory-infrastructure category. Zeroset’s bet is that long-horizon enterprise agents need process state, not just conversational memory. The distinction matters for any workflow that spans days or weeks and crosses system boundaries: claims handling, release approvals, multi-step procurement.

Gradient’s investment thesis, published the same day, makes the same point. Consumer agents succeeded because they trained on the public web. Enterprise agents still lack a representation of how work actually moves inside a private organization.

## What changes for agent stacks

If the state layer works at production scale, two things shift. First, agent harnesses stop shipping ever-larger context windows to compensate for missing structure. Second, enterprises can keep the operating model of their own workflows inside their own boundary instead of hoping a general model eventually invents it.

The numbers that matter right now are still early: $5.2 million, five people, closed preview, two public benchmarks. The claim is larger—that the bottleneck for enterprise automation is no longer model capability but the absence of a system of state.

[ Zeroset: Modeling the Enterprise for Long-Horizon Agents Gradient Ventures announces its investment in Zeroset and the Nebula state layer for enterprise agents. Gradient Ventures ](https://www.gradient.com/blog/posts/investing-in-zeroset/?ref=nexi.fund) 

[ Announcing Zeroset’s Pre-seed Official announcement of the $5.2M pre-seed and Nebula closed research preview. Zeroset ](https://zeroset.com/announcing-our-pre-seed?ref=nexi.fund) 

[ AI agents don’t always understand how companies work Business Insider covers Zeroset’s $5.2M raise, founding team, and competitive positioning against Mem0 and Zep. Business Insider ](https://www.businessinsider.com/zeroset-startup-funding-ai-agents-gradient-ventures-enterprise-2026-10?ref=nexi.fund)