Zeroset

A memory and state layer for autonomous AI agents in enterprise workflows.

Cover Block

From the public record

Field	Value
Name	Zeroset
Tagline	A memory and state layer for autonomous AI agents in enterprise workflows.
Headquarters	San Francisco, US [LinkedIn]
Founded	2025 [LinkedIn]
Stage	Pre-Seed [Business Insider, October 2026]
Business Model	SaaS
Industry	Deeptech
Technology	AI / Machine Learning
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (2), Akshat Kannan and William Zhang [Business Insider, October 2026]
Funding Label	Pre-seed
Total Disclosed	~$5,200,000 [Business Insider, October 2026]

The Short Version

PUBLIC Zeroset is building a memory and state layer for autonomous AI agents, and it merits attention now because it has moved from concept to a financed pre-seed effort around a specific enterprise problem: agents often lack a reliable record of how work actually unfolds across tools and teams [Business Insider, October 2026] [Zeroset LinkedIn]. The founding story is early but clear in the public record: co-founder and CEO Akshat Kannan reportedly started the company in January 2026 after leaving Stanford, while co-founder William Zhang had left the University of Texas in 2025 and lists his Zeroset co-founder start date as August 2025 on LinkedIn, which suggests the company formation timeline still needs tighter reconciliation [Business Insider, October 2026] [William Zhang LinkedIn].

Nebula, the company’s first product, connects to systems including Microsoft 365, SharePoint, Outlook, Teams, GitHub, and workplace messaging tools, then tracks how information, decisions, and activity change over time rather than simply retrieving static documents [Business Insider, October 2026]. That positioning matters because the wedge is not generic enterprise search, but temporal workflow memory: a persistent record agents can use to repeat prior successful processes and avoid repeating mistakes, with the product still in closed research preview as of October 2026 [Business Insider, October 2026] [zeroset.com].

The team is small, reportedly five employees at the time of Business Insider’s coverage, and the public evidence points to an engineering- and research-led buildout rather than a go-to-market-heavy organization at this stage [Business Insider, October 2026] [LinkedIn]. The founder profiles are nontraditional but relevant to an ambitious infrastructure bet, and the current hiring mix, including systems, distributed systems, and research roles, is consistent with a company still proving core architecture and deployment requirements before broad commercialization [LinkedIn] [bebee.com].

On financing, Zeroset disclosed a $5.2 million pre-seed round in October 2026, co-led by Gradient and 2048 Ventures with participation from Leblon Capital, which is substantial enough to fund a first technical team but does not by itself settle product-market fit [Business Insider, October 2026] [Crunchbase]. The planned business model combines software licensing with usage-based pricing tied to data volume and agent activity, which fits the product’s infrastructure posture, though pricing durability and deployment complexity remain open questions until named customers and production case studies emerge [Business Insider, October 2026]. Over the next 12 to 18 months, the key public markers to watch are whether Nebula exits preview, whether Zeroset can convert its integrations into repeatable enterprise deployments in manufacturing, supply chain, or financial research, and whether the company can show that workflow memory is a distinct budget line rather than a feature absorbed by broader agent platforms [Business Insider, October 2026].

Single-source, plausible -- Relies primarily on Business Insider with partial corroboration from LinkedIn, Crunchbase, and the company website.

Taxonomy Snapshot

Axis	Value
Stage	Pre-Seed
Business Model	SaaS
Industry / Vertical	Deeptech
Technology Type	AI / Machine Learning
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding	Pre-seed, total disclosed approximately $5.2 million [Business Insider, October 2026]

The Company in Brief

PUBLIC Zeroset presents as a very new San Francisco company, and the public record is still thin in the way early pre-seed infrastructure startups often are. Its Crunchbase profile identifies the company as Zeroset, based in San Francisco and founded in 2025, while the company website centers Nebula as its initial product and frames the business around memory infrastructure for AI applications rather than a broader services model [Crunchbase] [zeroset.com].

The chronology that can be supported from company-controlled and database sources is narrow but usable. LinkedIn and Crunchbase place the company’s founding in 2025, and the website now markets Nebula under the Zeroset name, indicating that the product identity was in place by the time of the captured site record [Crunchbase] [LinkedIn] [zeroset.com]. Crunchbase also lists a pre-seed financing of $5.2 million, disclosed in October 2026, which marks the first clear external milestone visible in the public company profile [Crunchbase].

What remains less settled from these sources is the formal corporate picture beyond the operating brand. No state filing or legal entity detail was provided in the source set used for this section, so the cleanest read is that Zeroset is a San Francisco AI infrastructure startup, founded in 2025, with Nebula as its first disclosed product and a pre-seed round recorded by October 2026 [Crunchbase] [zeroset.com].

Single-source, plausible -- This section relies primarily on Crunchbase and the company website, with partial corroboration from LinkedIn.

What They Have Built

MIXED

The product story is early, but the core claim is specific. Zeroset describes itself as an "applied AI research lab" building "the memory and state layer for autonomous agents," and public reporting frames Nebula as infrastructure for AI systems that need persistent state across tools, users, and time [LinkedIn] [TechCrunch, October 2025] [zeroset.com, Unknown]. Business Insider reported that Nebula connects to enterprise systems including Microsoft 365, SharePoint, Outlook, Teams, GitHub, and workplace-messaging tools, then tracks how information, decisions, and activity change over time rather than simply indexing static documents [Business Insider, October 2026].

That distinction matters because the company is not presenting Nebula as a general retrieval layer. According to Business Insider, the intended output is a workflow history that agents can reference to repeat successful procedures and avoid prior mistakes, with the product wedge centered on temporal workflow memory, meaning the sequence of events and decisions behind work [Business Insider, October 2026]. Nebula remained in a closed research preview as of October 2026, which suggests the technical thesis has attracted capital before broad market validation, and the planned commercial model was reported as a mix of license fees and usage-based pricing tied to data volume and agent activity [Business Insider, October 2026].

The public technical surface also hints at the underlying architecture, although some of that must be treated as directional rather than confirmed product documentation. An observed Systems (Rust) job posting referenced Nebula's state engine, hierarchical vector graph, provenance, and storage across in-memory cache, local NVMe, and S3-compatible object storage, while a distributed systems posting referenced object-storage-native architecture, distributed execution, partitioning, multi-region deployment, and customer-managed infrastructure (inferred from job postings) [LinkedIn] [bebee.com]. Separately, a GitHub reference indicates automated SDK regeneration for Nebula, which is consistent with an integration-heavy platform, but does not by itself establish customer usage or production maturity [GitHub, 2026].

Unconfirmed -- This section combines independent press reporting with company-controlled sources and architecture inferences from job postings.

Market Size and Demand

PUBLIC

The market matters now because enterprises are moving from single-step copilots toward longer-running AI agents, and those systems break quickly when they cannot retain workflow context across tools and time [Business Insider, October 2026] [TechCrunch, October 2025].

Zeroset is not entering a neatly bounded software category with a cited standalone market size in the available public record. What the evidence does support is a narrower functional demand signal: buyers want AI systems that can do more than retrieve documents, especially in workflows where decisions unfold across email, chat, documents, and code systems over time [Business Insider, October 2026]. Zeroset's pitch for Nebula, a memory and state layer for autonomous agents, sits inside that shift from static retrieval toward persistent workflow memory [Business Insider, October 2026] [zeroset.com].

That framing matters because the company's stated initial users are enterprises and AI-native companies operating long-running agents, with early focus on manufacturing, supply-chain operations, and financial research [Business Insider, October 2026]. Those are not arbitrary verticals. They are settings where state changes, handoffs, and historical decisions affect current work, which makes temporal memory more valuable than a simple document index, at least in theory [Business Insider, October 2026]. The public evidence does not establish market size by vertical, so the safer read is that Zeroset is pursuing a cross-functional infrastructure layer rather than a single vertical application [Business Insider, October 2026] [TechCrunch, October 2025].

The main demand driver in the cited reporting is architectural, not cosmetic. Nebula is described as connecting to systems such as Microsoft 365, SharePoint, Outlook, Teams, GitHub, and workplace messaging tools, then tracking how information, decisions, and activity evolve over time [Business Insider, October 2026]. If enterprises continue deploying agents into operational workflows, products that preserve state across fragmented systems should benefit from the same adoption curve, although that remains a conditional inference rather than a demonstrated outcome for Zeroset specifically [Business Insider, October 2026].

Adjacent markets are easier to identify than the precise core category. Nebula touches enterprise search, knowledge management, agent infrastructure, workflow automation, and data integration, but the company's claimed differentiation is that it stores sequence and state, not just content or embeddings [Business Insider, October 2026] [zeroset.com]. That means Zeroset may compete for budget against retrieval layers and orchestration tools even if it does not describe itself that way, because procurement lines for early agent systems are often bundled under broader AI infrastructure or workflow software initiatives [TechCrunch, October 2025].

A small public sizing table is the most honest way to present the market evidence here, because the available sources provide category direction but not a sourced TAM model for Zeroset's exact segment.

Market claim	Relevance to Zeroset	Source
AI agents need workflow memory beyond static document retrieval	Directly aligned with Nebula's core product thesis	[Business Insider, October 2026]
Memory infrastructure for long-horizon agents is emerging as a distinct need	Supports category formation around persistent state for agents	[TechCrunch, October 2025]
Initial target environments include manufacturing, supply chain, and financial research	Suggests early wedge in process-heavy, history-dependent workflows	[Business Insider, October 2026]

from the public record is modest but useful. The category signal is real, but it is still pre-taxonomy: investors can underwrite a credible problem statement before they can underwrite a clean market map.

Regulatory and macro forces cut both ways. On the positive side, enterprise caution around hallucinations, auditability, and workflow reliability may favor systems that retain provenance and decision history, since those features can make agent behavior easier to inspect [Business Insider, October 2026]. On the limiting side, the same enterprise environments, especially in manufacturing, supply chain, and financial research, tend to have longer security review cycles and higher integration burdens, which can slow adoption for any agent infrastructure vendor connecting across core systems [Business Insider, October 2026]. Nebula's closed research preview status also suggests that the market is still early enough that product readiness and trust, not just demand, will shape near-term expansion [Business Insider, October 2026].

Single-source, plausible -- This section relies primarily on Business Insider's October 2026 reporting, with partial category corroboration from TechCrunch and company materials.

Who Else Is Fighting for This

Market Position

MIXED Zeroset is positioning itself less as another retrieval layer and more as workflow memory infrastructure for enterprise agents, which puts it adjacent to knowledge indexing tools and systems-integration platforms rather than in direct overlap with traditional enterprise search alone [Business Insider, October 2026] [zeroset.com] [LinkedIn].

The public record is thin on named direct rivals, so the cleaner way to map the field is by substitute category rather than by logo. On one side sit incumbent systems of record, Microsoft 365, SharePoint, Outlook, Teams, GitHub, and workplace messaging tools, which already hold much of the operational exhaust Nebula aims to interpret, but were not designed to preserve cross-tool causal history for agent execution [Business Insider, October 2026]. On another side sit generic retrieval and document-indexing layers, which can surface files or messages but, according to Business Insider's description of Nebula, do not necessarily track how decisions and activities evolve over time across workflows [Business Insider, October 2026]. A third substitute category is internal engineering effort: large enterprises and AI-native companies can assemble their own state and memory stack around existing data stores and orchestration tools, particularly if they already control the surrounding agent runtime.

That framing matters because Zeroset's wedge is narrow but legible. Business Insider described Nebula's core claim as temporal workflow memory, preserving sequence and state across enterprise systems so agents can replay what worked and avoid prior errors [Business Insider, October 2026]. If that works in production, the company is not competing on raw model quality or broad app surface. It is competing on whether its memory layer becomes a hard-to-replace substrate between enterprise data exhaust and long-running agents.

The company's edge today appears to rest on product framing, early technical focus, and investor signaling, but each of those advantages is still perishable. Public materials show a product in closed research preview, a five-person team at the time of coverage, and active hiring across systems, distributed systems, and research roles, which is consistent with a company trying to build difficult infrastructure before broad go-to-market expansion [Business Insider, October 2026] [LinkedIn] [bebee.com]. The $5.2 million pre-seed co-led by Gradient and 2048 Ventures gives Zeroset capital and some credibility with technical buyers, yet that is not the same as distribution, and no public customer list or reference deployments have been named [Business Insider, October 2026]. In practical terms, the edge is durable only if Nebula accumulates proprietary workflow-state data or embedding into customer systems that would be costly to reproduce. Based on the public record, that durability is still a hypothesis rather than an observed fact.

The main exposure is control of the surrounding stack. Microsoft already owns several of the systems Nebula connects to, and GitHub sits inside a broader developer workflow that could itself become a memory source or agent-control plane, which leaves Zeroset vulnerable if platform owners decide temporal state should be native rather than third-party [Business Insider, October 2026]. There is also a build-versus-buy risk at the upper end of the market: sophisticated enterprises may prefer to keep memory, provenance, and agent-state infrastructure inside their own environment, especially where data governance is sensitive. Public job listings suggest Zeroset is at least aware of that objection, with references to customer-managed infrastructure and object-storage-native architecture, but those postings do not establish shipped capability at scale [bebee.com] [LinkedIn].

Over the next 18 months, the most plausible competitive scenario is a split market. Microsoft is the winner if enterprise buyers decide acceptable agent memory must come bundled with the productivity suite and security perimeter they already run, because that would favor incumbency and distribution over a specialist layer. Zeroset is the relative winner if long-horizon agents prove to need cross-system workflow memory that incumbent collaboration and storage products still do not model well, and if early deployments turn that technical distinction into a repeatable enterprise sale [Business Insider, October 2026]. In the opposite case, the likely loser is not necessarily a named startup rival, but the independent middleware category itself: if customers settle for retrieval plus orchestration, specialist memory layers will have a narrower window to matter.

Unconfirmed -- This section relies primarily on one independent news report for Zeroset's positioning and product scope, plus company and job-posting materials for supplemental context [Business Insider, October 2026] [zeroset.com] [LinkedIn] [bebee.com].

Opportunity

Upside Case

PUBLIC The prize here is not another enterprise AI feature, it is a control point: if Zeroset becomes the system that gives long-running agents reliable memory across enterprise tools and workflows, it could sit underneath a meaningful share of agentic software spend as that market moves from demos into production [Business Insider, October 2026] [TechCrunch, October 2025] [zeroset.com].

The headline opportunity is straightforward. Zeroset is trying to supply the persistent state layer that agents need once work extends across users, systems, and time, rather than a single prompt or static document set [Business Insider, October 2026] [zeroset.com]. That is a narrower claim than "enterprise AI platform," but it can still become a large one if the problem is real and recurring. The public evidence suggests management is aiming at exactly that bottleneck: Nebula connects to systems such as Microsoft 365, SharePoint, Outlook, Teams, GitHub, and workplace messaging tools, and tracks how information, decisions, and activity evolve over time so agents can repeat successful procedures and avoid prior mistakes [Business Insider, October 2026]. If enterprise buyers conclude that retrieval alone is insufficient for long-running agent workflows, a memory-and-state layer becomes infrastructure rather than application software, which is the part of the stack that can travel across many use cases.

The growth paths are still early, but they are legible from the product design and stated customer focus. Zeroset says its initial buyers are enterprises and AI-native companies operating long-running agents, with early focus on manufacturing, supply-chain operations, and financial research [Business Insider, October 2026]. Those are all settings where process history matters, where the cost of repeated mistakes can be high, and where systems sprawl makes durable context harder to maintain.

Scenario	What happens	Catalyst	Why it's plausible
Workflow memory standard	Nebula becomes the default memory layer for enterprise agent deployments that span collaboration, code, and operational systems	Closed research preview converts into successful early enterprise deployments, proving that temporal workflow memory improves agent reliability in production [Business Insider, October 2026]	Nebula is already framed around persistent workflow history across Microsoft 365, SharePoint, Outlook, Teams, GitHub, and messaging systems, which puts it close to day-to-day enterprise work rather than a narrow single-app use case [Business Insider, October 2026]
AI-native infrastructure supplier	Zeroset becomes the embedded backend for startups building long-horizon agents, selling memory and state as a core primitive	Productization of the planned license plus usage-based model tied to data volume and agent activity [Business Insider, October 2026]	The product is positioned as memory infrastructure for long-horizon agents and persistent state across tools, users, and time, which fits an API or infrastructure motion if developers adopt it early [TechCrunch, October 2025] [zeroset.com]
Regulated workflow wedge	The company wins in domains where auditability of decisions and process history matter, then expands from there	A category-tipping deployment in financial research or complex operations where buyers value historical provenance as much as raw model performance [Business Insider, October 2026]	Zeroset's public framing emphasizes how decisions and activity change over time, not just document indexing, which is directionally aligned with environments where sequence and provenance matter [Business Insider, October 2026]

What compounding could look like is also fairly clear. If Nebula is integrated once into the systems where a company's work already happens, each additional agent or workflow can run on the same memory substrate, which should make expansion easier than a point solution sale [Business Insider, October 2026]. The product description hints at that dynamic: the more workflow history the system captures across tools and over time, the more useful its state layer becomes for future automations, and the harder it is to swap out without retraining procedures or losing organizational memory [Business Insider, October 2026] [zeroset.com]. The hiring pattern points in the same direction. Zeroset has posted roles in systems, distributed systems, and research, including work tied to state engines, provenance, storage, and distributed execution, which suggests the company is building foundational infrastructure rather than a thin application wrapper [LinkedIn] [bebee.com].

The size of the win is best framed through infrastructure outcomes, not current traction. No close public comparable is confirmed in the source set, so precision would be false here. Still, if the "workflow memory standard" scenario plays out and Nebula becomes a common dependency for enterprise agent deployments, the result could support multi-billion-dollar enterprise value, because infrastructure that sits across many workflows and prices on license plus usage can participate in both seat expansion and workload growth (scenario, not a forecast) [Business Insider, October 2026]. That upside remains conditional on two things the public record does not yet prove: first, that enterprises will treat memory and state as a distinct budget line, and second, that Zeroset can move from research preview to repeatable production adoption. The opportunity is large because the claimed bottleneck is foundational, not because scale has already been demonstrated.

Single-source, plausible -- Built primarily from one named-publisher report, with partial corroboration from company website, LinkedIn, and job postings.

Sources