Tell us how your team runs AI agents and what your environment requires — routing and capacity, secret handling, identity integration, and the servers or blocks you depend on. Enterprise capabilities are scoped against your setup rather than provisioned automatically.
Discuss dedicated routing and capacity for your team’s workload. Scoped per engagement rather than provisioned automatically by submitting this form.
Local credentials are held in an encrypted keyring under ~/.steroid and injected at execution. Connected mode forwards the credentials a tool needs to run. Organisation-wide vault and audit requirements are scoped with you.
Accounts today use email/password with JWTs, plus separate connection keys for clients. SAML and OIDC are not yet implemented — tell us what your identity provider requires.
Priority feature requests and direct engineering access. If a server, block, or integration your team depends on is missing, it gets built against your use case rather than queued behind a public roadmap.
// For faster responses, please include your phone number.
Enterprise is an inquiry channel rather than a self-serve tier. Dedicated routing, organisation-wide secret handling, audit requirements, and identity integration are scoped against your environment during the conversation. Submitting this form starts that scoping — it does not by itself provision dedicated infrastructure.
Not today. Accounts currently authenticate with email and password and are issued JWTs, while connected clients use a separate connection key that is shown once and stored only as a SHA-256 hash, with a one-hour overlap on rotation. SAML and OIDC are not implemented. If your rollout depends on an identity provider, say so in the form so it can be scoped explicitly.
In Local mode, credentials sit in an encrypted keyring under ~/.steroid, cached for 15 minutes by default and invalidated when inventory changes. In Connected mode, the cloud MCP layer receives the credentials a given tool needs in order to execute it, and flows you build persist their own encrypted connections. Credential values are supplied at the tool boundary rather than being pasted into prompts, but this is not an unconditional guarantee of zero retention or organisation-level audit.
Local mode exposes three operations: recall_memory, recall_turn and recall_diff. Connected mode exposes nine, adding search_tools, run_tool, search_workflows, run_workflow, create_workflow and repair_workflow. The size of the server catalogue does not change that count, because servers are reached through search_tools and run_tool rather than being registered individually — so context cost stays flat as the catalogue grows. Measure your own usage rather than assuming a fixed saving.
Discovery issues a short-lived capability handle — a process-local value with a default five-minute lifetime — and execution checks it against the workflow, step and candidate it was issued for. An agent cannot execute a tool outside that sequence. This constrains which calls are eligible; it is not a guarantee that the model chose the semantically correct tool.
Setup targets Cursor, Claude Code, Codex and GitHub Copilot. Installation, mode selection, IDE wiring and Connected credential setup are distinct steps with their own prerequisites, so verify each with the built-in setup diagnostics rather than assuming a single command finishes all of them.
Yes. A flow can begin with a webhook trigger and end with a response node, making it callable from systems you already run, and it can chain MCP server blocks with LLM nodes in between. Note that webhook authentication defaults to none and must be configured explicitly, and that the published schema is metadata rather than automatic runtime validation of incoming payloads.