Superblocks: AI-augmented internal tools platform; on-premise agent, Clark AI in beta

Superblocks is a platform for building internal apps, backend APIs, and scheduled jobs. The visual builder handles the UI layer (tables, charts, forms, maps — 100+ components); integrations connect to 50+ databases, data warehouses, and SaaS APIs; the AI layer generates and explains code across SQL, Python, JavaScript, and JSON. An on-premise execution agent keeps data traffic inside your network.

The on-premise agent split

Superblocks runs on a hub-and-spoke architecture. The Superblocks cloud hosts the control plane — app definitions, permission policies, routing. An on-premise agent (open-source, deployed via Docker on AWS, GCP, Azure, or any on-premises data center) executes API calls and database queries inside your network perimeter. Credentials for Snowflake, Postgres, internal APIs, and other integrations live on the agent; they are never transmitted to Superblocks' servers.

The agent is stateless by design: state lives in the database connections it proxies. This means horizontal scaling and crash recovery are straightforward — restart the container and reconnect. The tradeoff is that all execution latency passes through your deployment; teams on slow internal networks will see higher round-trip times than teams using pure cloud execution.

The agent source is at github.com/superblocksteam/agent. Teams can audit dependencies and confirm what the execution layer does before deploying. Superblocks frames this as a security property: a customer's security team can review the agent code rather than accepting vendor trust.

Integrations and query authoring

50+ integrations at current count: PostgreSQL, MySQL, MongoDB, Snowflake, Databricks, Google BigQuery, Amazon S3, and the major SaaS APIs. Query blocks run SQL natively; Python and JavaScript blocks run in the agent sandbox; REST API blocks support custom headers and auth schemes. The AI query assistant generates SQL or code from a natural-language description of the intent and explains existing queries on demand.

The AI generation layer passes the query description and schema metadata to the model; the generated code drops into the editor for review before execution. There is no silent execution of AI-generated queries.

Clark AI: multi-agent architecture, private beta as of May 2025

In May 2025 Superblocks announced Clark — a multi-agent system for building complete internal apps from natural-language prompts. The architecture runs five specialized agents in parallel: Design (layout and component selection), IT (integration wiring), Engineering (business logic), Security (policy enforcement), and QA (test generation and validation). A user describes what they want; the agents assemble a working app that already has RBAC, audit logging, and the organization's design system applied before it ships.

Clark entered private beta on May 27, 2025. Teams plan ($100/AI Builder/month, annual billing) includes Clark. Enterprise adds Clark Integration Memory and Clark Organization Memory — persistent context about how your org connects systems and what your policies require, applied automatically on each build.

Clark is distinct from the 2023 AI query assistant. The query assistant helped developers write individual queries faster within an existing app. Clark is intended to eliminate the requirement for a developer on the standard internal-tool build path.

Pricing

Teams: $125/AI Builder/month (monthly) or $100/AI Builder/month (annual). 14-day free trial, self-serve. Includes 100 AI Credits per AI Builder monthly, 50+ integrations, staging and production environments, RBAC, and 1 hosted app. Additional apps cost $100/month each. Scales up to 15 Builders.

Enterprise: custom pricing. Adds VPC deployment (Hybrid or Cloud-Prem on AWS/GCP/Azure), SSO/SAML/OIDC, Secrets Management (Hashicorp Vault, cloud providers), Embedded Applications, unlimited end users, audit logs, and an observability pipeline. Requires a demo booking.

Retool: the relevant comparison

Retool is the dominant internal tools platform. Airplane (the other commonly named competitor) was acquired by Airtable in 2023 and wound down; it is not an active product as of 2026.

Against Retool the comparison axes that matter are deployment architecture, AI depth, and pricing model. Superblocks uses an on-premise agent by default for enterprise workflows — Retool offers a similar on-premises option but structures it differently. The Clark AI multi-agent system (private beta 2025) has no direct equivalent in Retool's current product. Retool's pricing starts below the Superblocks Teams tier for individual builders but scales with seat count and additional apps in ways that differ by contract.

Field notes

Developer in the Show HN thread praised the AI query generation UX as "very cohesive" — specifically the diff preview before pasting, inline method editing, and subtle animations. Noted it felt more considered than standard LLM integrations of the time. [community-thread, 2023-06-27, HN item 36495680]
Same thread: a developer abandoned a webhook receiver integration when the sending service used a non-standard authentication scheme. Superblocks was "rigid" — the platform could not match the sender's custom auth. Custom incoming-webhook auth required workarounds outside the platform at that point. [community-thread, 2023-06-27, HN item 36495680]
Clark multi-agent system announced with $60M total funding (Series B: $23M led by Spark Capital, Kleiner Perkins, Meritech Capital, Greenoaks). Private beta May 27, 2025. [dev-post, 2025-05-27]

Sources: community-thread — HN item 36495680 (2023-06-27, Show HN: Superblocks AI – AI coding assistant for internal apps, 114 points); dev-post — superblocks.com/blog/announcing-clark-ai (2025-05-27); Superblocks pricing page observed 2026-05-20.