Chalk

Chalk is a data platform for machine learning and AI inference. Its core product, the Context Engine, is a real-time feature computation engine in which developers define features, embeddings, LLM outputs and prompts once in Python and Chalk computes and serves them from underlying data sources at inference time. The platform is designed to keep training and serving consistent (point-in-time correct queries, replay of any computed value), and includes built-in monitoring, drift and data-quality observability, and a branch-based deployment model for feature versioning. A second product line, Chalk Compute, provides governed, sandboxed execution environments for agents and remote processes: sandboxes are isolated with gVisor, receive their own scoped cloud identity, boot in under a second, and are described as scaling to 10,000 isolated instances in under 10 seconds; the same SDK can run persistent GPU-backed model servers.

Chalk emphasizes deployment inside the customer's own cloud/VPC under the customer's IAM roles, using the customer's existing database as the online and offline store rather than bespoke storage. Its real-time serving engine scales horizontally and executes complex queries on a Rust-based runtime; the company advertises 100,000 QPS at under 5ms and single-digit-millisecond feature serving. Developers install the SDK via pip (chalkpy for the Context Engine, chalkcompute for agents and models), write resolvers, deploy with `chalk apply`, and query features in real time or backfill them as of any past point. Chalk Compute runs on AWS or GCP, while the Context Engine also supports Azure. Workflows highlighted by the company include fraud and risk (withdrawal limits, fraud scoring), credit (computing income from Plaid transactions), caching controls, and predictive maintenance from streaming device data.

The company positions the platform for real-time inference workloads originally common in fintech—compliance, KYC, fraud detection, credit underwriting—and has extended into identity verification, healthcare, e-commerce, marketplaces and recommendations. Chalk maintains offices in San Francisco (55 Stockton St., Floor 4) and New York (54 W 21st Street, 401).

Founding story

Chalk's three co-founders are repeat founders with backgrounds in payments, fintech risk and large-scale data infrastructure. CEO Marc Freed-Finnegan spent several years at Google, where he helped launch the first version of Google Wallet, and then founded Index, which Stripe acquired and turned into Stripe Terminal. Elliot Marx began his career at Affirm, where he built the early risk and credit data infrastructure system that the company cites as the direct inspiration for Chalk, and later co-founded Haven Money, acquired by Credit Karma to power its banking products. Andy (Andrew) Moreland worked at Palantir on large government data infrastructure projects and co-founded Haven Money with Marx. According to a secondary profile, the founders started Chalk around 2022 to systematize the real-time ML infrastructure pioneered inside fintech into an industry-agnostic platform, motivated by the limits of batch-processed data for real-time decisions. The company name references the chalkboard used by mathematicians, and its logo derives from the QED symbol marking the end of a mathematical proof.

Business model

Chalk sells an enterprise software platform deployed into the customer's own cloud account (AWS, GCP, and for the Context Engine also Azure), rather than as a hosted data store. A secondary profile characterizes it as a platform-as-a-service model for enterprise data infrastructure. Customers install Python SDKs, connect their existing databases and data sources, and use Chalk for feature computation, serving, observability and agent sandboxing.

Sources do not state pricing or contract structure; engagement begins with a booked demo, indicating an enterprise sales motion for platform access.

Traction

The company reports serving customers including Doppel, Sunrun, Whatnot, Socure, Found, Medely and iwoca, with Whatnot cited as computing more than 300 million features per second for real-time recommendations, Mission Lane using Chalk for real-time credit approvals and fraud detection, Apartment List for dynamic search recommendations and MoneyLion for unified ML development. A customer executive quoted on the site describes moving hundreds of millions of features per second at roughly 1MB payloads with P99 latency of 100ms. Headcount grew from about 30 to about 60 in the year preceding November 2025; another profile cites roughly 45 employees. Recognition includes the CB Insights AI 100 (2024) and a Fast Company World's Most Innovative Companies feature (March 2026).

Latest developments

In May 2025 Chalk announced a $50M Series A at a $500M valuation led by Felicis, with Triatomic Capital and existing investors General Catalyst, Unusual Ventures and Xfund participating, and Aydin Senkut of Felicis joining the board; the company said proceeds would fund a more general compute framework. Subsequent activity includes an August 2025 quarterly product update, SciPy 2025 participation, the launch of Chalk Compute for governed agent sandboxes and model servers alongside the Context Engine, a webinar series on building a real-time context layer, a June 2026 customer post on Melio's build-vs-buy decision, and a March 2026 Fast Company innovation feature.