SyntheticBrew Cloud Quick Start

SyntheticBrew Cloud is the managed path: SyntheticBrew operates the runtime and database while you configure agents and connect your application. By the end of this guide you will have a chat-enabled schema with a tested entry agent.

Before you begin

Create an account at syntheticbrew.ai/register and open the Admin Dashboard. A schema is the product-facing workflow boundary: it selects one entry agent, contains its delegation relationships, and controls whether the chat endpoint accepts requests.

Cloud may provide a platform model for initial testing. Configure your own chat model when you need a specific provider, model, endpoint, or billing account.

1. Understand the starting schema

Open Schemas. New workspaces include a starting schema; you can open it or choose New Schema to fork a template. A template creates independent agents and delegation relationships that you can edit.

In the schema detail page, confirm the entry agent and turn Accept chat requests on. The expected result is a schema that can receive dashboard, widget, and REST chat traffic.

2. Configure a model when needed

Open Models and check the Chat tab. If no platform default is available, or you want your own provider, select Add Model and set:

Kind: Chat. A model’s kind cannot be changed after creation.
Provider and model name: choose a registry entry or enter an OpenAI-compatible model.
Base URL: required for Ollama and custom OpenAI-compatible endpoints.
API key: stored for this model; it is not returned by the API after saving.

Use Verify before assigning the model. For per-request credentials instead, enable BYOK and send the documented BYOK headers.

3. Create or refine the agent

Open Agents, select New Agent, and give it a display name, URL-safe name, system prompt, lifecycle, and chat model. Start with a narrow responsibility and explicit boundaries.

Return to Schemas, open the schema, and make the agent the entry agent if it is not already. Add delegation relationships only when a specialist has a distinct job. Delegation is derived from schema relations; it is not a writable can_spawn list on the agent.

4. Test the real workflow

Open the schema and expand the Test Flow panel at the bottom of Admin. Ask a request that exercises the intended behavior and watch the structured events, tool calls, and delegation steps. Continue in the same session to verify context behavior.

The expected result is a completed response from the configured entry agent. If the request fails:

verify the assigned chat model;
open Tool Call Log for tool errors;
inspect MCP Servers and Resilience for connection or circuit-breaker state;
check that Accept chat requests is enabled on the schema.

5. Ground or connect the agent

Choose the data surface that matches the job:

Knowledge for prose in TXT, Markdown, CSV, PDF, or DOCX files. Create a base with an embedding model, upload files, wait for ready, link the base to the agent, and ensure the Knowledge capability is enabled.
Knowledge Graphs for typed entities that must be filtered or retrieved exactly, such as rooms, prices, catalog records, and IDs.
MCP Servers for live systems and actions. Add a catalog or custom server, verify discovered tools, then attach only the tools the agent needs.

Re-test the schema after each binding so the result is attributable to one change.

6. Connect a product or coding agent

For a product UI, use the REST and SSE integration or generate a credential-safe widget embed. For automated provisioning, follow Connect a Coding Agent; OAuth is the default, and manual scoped keys are reserved for headless or otherwise unsuitable OAuth environments.