Zep AI recruits an FDE chief to get agent memory past security review

Why it matters

Zep AI's hire shows where agent-memory startups encounter their hardest scaling problem: private-cloud deployments, security reviews and integrations that cannot remain founder-managed forever.

Daniel Chalef (@danielchalef) is recruiting a Head of Forward Deployed Engineering for Zep AI, putting a senior operator in charge of carrying its agent-memory infrastructure from architecture meetings through 90 days in production.

The open role pays $220,000 to $270,000 and offers 1.20% to 1.75% equity. The remit covers integration code, reference architectures, deployment tooling, security reviews, customer infrastructure and hiring the engineers who will eventually form Zep AI's forward-deployed team.

That is a consequential hire for a small business. Y Combinator lists Zep AI as a San Francisco-based member of its Winter 2024 batch, founded in 2023. Chalef previously founded KnowledgeTree, then worked across marketing, data science and corporate development at Domino Data Lab and SparkPost. Zep AI describes him as an engineer and former head of machine learning at SparkPost.

The background fits the assignment. Chalef has spent his career moving between building enterprise software and getting enterprises to buy it. Zep AI's new engineering chief will operate at that same boundary, with code, contracts and deployment deadlines arriving in the same meeting.

The hire is part of the product strategy

Forward-deployed engineering roles tend to appear when a technical product works in controlled conditions and then encounters the peculiarities of large customers: identity systems, encryption policies, procurement requirements, cloud restrictions and security teams with their own calendars.

Zep AI states the problem plainly in the listing: getting its software into customer environments is an engineering problem of its own. The new head will own technical outcomes for large accounts from the first architecture conversation through production, including deployments inside customer-controlled cloud infrastructure.

The job requires fluency in AWS, Kubernetes and Terraform, along with Python and either Go or TypeScript. Zep AI also wants someone who has shipped a substantial agent system, led customer-facing technical staff and continued writing production code.

Chalef is hiring this person to reduce the custom work required by each deployment. The listing instructs the future hire to turn work completed twice into either a reference architecture or a product requirement. Success will be measured through the time between signing and production, as well as the amount of custom engineering consumed by each account.

That mandate keeps Zep AI from drifting into open-ended consulting as enterprise requests accumulate. It also gives Chalef a feedback loop from live deployments into the product roadmap. The forward-deployed group is expected to discover recurring problems, write the first fixes and hand reusable lessons back to core engineering.

Agent memory meets enterprise infrastructure

Zep AI sells what it calls a Context Lake, a system that combines conversations, documents, events and business data into temporal context graphs. Those graphs are meant to help agents retrieve relevant information while preserving when facts changed, where information came from and which users or applications may access it.

The architecture grew from Graphiti, Zep AI's open-source framework for temporal knowledge graphs. A 2025 research paper authored by Preston Rasmussen, Pavlo Paliychuk, Travis Beauvais, Jack Ryan and Chalef described a memory layer that combines conversational history with structured business data. It tracks relationships over time instead of treating retrieved text as permanently current.

Zep AI says its commercial service supports millions of context graphs per deployment with retrieval below 200 milliseconds. It offers managed cloud, customer-controlled encryption keys and bring-your-own-cloud deployments inside a customer's virtual private cloud. The last option pushes Zep AI directly into the security, networking and compliance work covered by the new role.

Zep AI also claims more than 240 customers and 50% month-over-month annual recurring revenue growth on its YC listing. The listing gives no ARR figure or measurement period for that growth rate. Zep AI names Samsung, Zscaler, Twin Health and HoneyBook among its customers, alongside unnamed Nasdaq-100 and Fortune 500 technology businesses.

Even Zep AI's open-source adoption figures require careful reading. Its careers page describes Graphiti as having more than 30,000 GitHub stars, while the company description within the YC listing says more than 24,000. Both figures come from Zep AI-controlled hiring material and illustrate how quickly repository metrics can age.

Memory is becoming its own infrastructure market

Agent-memory infrastructure has attracted a growing field of competitors. Mem0 said in October 2025 that it had raised $24 million across its seed and Series A rounds. Letta has built an agent runtime around explicit memory and editable state, while LangChain's LangMem SDK gives developers tools for semantic, episodic and procedural memory inside the LangGraph orbit.

Zep AI is leaning into temporal graphs, governance and private deployment. That positioning raises the cost of serving each major customer because the software has to coexist with established security controls and data boundaries. The forward-deployed engineering hire is Chalef's answer to that cost: put an experienced builder beside the customer, then convert the resulting work into infrastructure that the next deployment can reuse.

The compensation and equity range show how much authority Chalef intends to place in the role. The new hire will define which accounts receive embedded support, help determine how engagements are scoped and priced, and build the function from scratch. Zep AI expects the person to enter a live deployment immediately, own every active deployment within 60 days and produce the first engagement model within 90 days.

For Chalef, the hiring decision moves a founder-led deployment process toward an operating system that another leader can run. Zep AI can keep winning technically demanding enterprise work only if each installation teaches the product how to require less installation the next time.