Chalk
Chalk is a production AI platform that combines data context, model runtime, and tools for developing and assessing models and agents.
What Chalk is
Chalk is a production AI platform that combines data context, model runtime, and tools for developing and assessing models and agents. Its feature store and federated database engine can retrieve features from streams, databases, and APIs at query time, making connected data available during inference. Runtime options include model serving, GPU workloads, agent sandboxes, and a model gateway supporting multiple provider types and OpenAI-compatible endpoints. Teams can also use evaluations, traces, notebooks, and fine-tuning. Chalk can be deployed in AWS, GCP, or Azure, with self-hosted data-plane workloads running inside the customer's cloud; a Chalk-hosted deployment is also described. Its agent sandboxes use gVisor isolation and can limit outbound network access by hostname, CIDR, and port. Chalk identifies data engineers, MLOps teams, AI/ML engineers, data scientists, and AI platform engineers as audiences. The company announced a SOC 2 Type I audit in 2023, with the report available by request. Pricing is available on request.
Good
- Feature store and federated engine resolve data at query time.
- Runtime includes serving, GPU workloads, sandboxes, and a model gateway.
- Evaluation tools include traces, notebooks, and fine-tuning.
- Self-hosted data-plane workloads run inside the customer's cloud.
- Sandbox outbound access can be restricted by hostname, CIDR, and port.
Watch out
- Paid product; pricing is available on request.
- SOC 2 Type I report is available only by request.
Who it suits
It is aimed at data and AI teams building or operating models and agents in production. The listed audiences include data engineers, MLOps teams, data scientists, and AI platform engineers.
Verdict
Chalk combines inference-time data context with model runtime and development tools. Teams should weigh its available deployment options and request pricing and the SOC 2 Type I report directly.
The facts
Each one read from Chalk's own pages, with the date it was checked.
The company
- Company location
- Chalk lists offices in San Francisco and New York.checked 3 Oct 2026
- Headquarters
- San Francisco, California, United Stateschecked 28 Sep 2026
What it does
- Compliance
- Chalk announced completion of a SOC 2 Type I audit in March 2023 and says the report is available by request.checked 2 Oct 2026
- Data sources
- Chalk says it can resolve features using data from streams, databases, and APIs at query time.checked 3 Oct 2026
- Evaluation
- Chalk lists evals, traces, notebooks, and fine-tuning among its tools for testing and improving models and agents.checked 3 Oct 2026
- Feature store
- Chalk lets teams define features and manage versions, then serve computed context at inference time.checked 3 Oct 2026
- Integrations
- The CLI documentation lists integration kinds including PostgreSQL, Snowflake, BigQuery, Kafka, MySQL, Redshift, and Databricks.checked 3 Oct 2026
- Intended users
- Chalk presents use cases including fraud detection, payments, underwriting, recommender systems, search and ranking, and dynamic pricing.checked 3 Oct 2026
- Learning tools
- The platform includes evaluations, traces, notebooks, and model fine-tuning.checked 2 Oct 2026
- Live context
- Its feature store and federated database engine serve context from connected data sources at inference time.checked 2 Oct 2026
- Managed security
- The self-hosted comparison says Chalk-hosted deployments are managed by Chalk with fine-grained access controls and SOC 2 Type II, ISO 27001, and GDPR compliance.checked 3 Oct 2026
- Model providers
- The Model Gateway supports provider kinds including OpenAI, Anthropic, Gemini, Vertex, Bedrock, Cohere, Mistral, and OpenAI-compatible endpoints.checked 3 Oct 2026
- Product
- Chalk describes its data platform as providing building blocks for machine learning, with an experience for developers.checked 3 Oct 2026
- Runtime
- The platform includes model serving, GPU compute, sandboxes for agents and code, and a model gateway.checked 3 Oct 2026
- Sandbox security
- Chalk says agent sandboxes use gVisor isolation and allow customers to restrict outbound network access by hostname, CIDR, and port.checked 2 Oct 2026
- Self-hosting
- Chalk offers self-hosted deployment, where data, features, and models stay in the customer’s infrastructure; the page says it supports fully air-gapped deployments.checked 3 Oct 2026
- SOC 2 report
- Chalk’s 2023 announcement says it completed a SOC 2 Type I audit and customers can contact [email protected] for a copy of the report.checked 3 Oct 2026
- Support
- Chalk’s startup program offers accepted teams hands-on support from forward-deployed engineers.checked 3 Oct 2026
