Agentic infra

The high-level system design of the agentic infra - who starts a run, the one run start, the runtime, what a run can do and know, and where it is stored.

1 min read Updated Oct 15, 2026

Agentic infra

This page is the map of the agentic infra. It shows every module and the path from a request to a result. Each module has its own page, and the numbers on the diagram match the module pages.

A person or a machine asks for work. Every run starts at Runs, and one runtime executes it. A run reaches the world only through capabilities and tools, and governance and observability apply at every stage. Dashed means not ours, not built yet, or async.

The path of one request

A person sends a chat message from a surface, or works in the product UI.
The Front Door selects one Agent or Workflow and starts a run. A product action and a matched trigger start a run directly.
Runs admits the run, freezes the published definition, and dispatches it to the runtime.
The runtime executes the steps. An agent step calls the model provider.
Steps call tools. Tools read and write context, memory, product state and storage.
The run reports progress and the result back to the surface that asked.

Modules

#	Module	What it owns
01	Surfaces and channels	Where work is asked for: the web app, the ac CLI, the channel gateway, and the planned MCP server.
02	Triggers	How events, webhooks and schedules match one rule and start a run without a person.
03	Front Door	How a conversation reads scoped context and answers, asks, or selects one Agent or Workflow.
04	Runs	The one run start: RunManager, admission, the frozen definition, and dispatch.
05	Runtime	The one runtime: published definitions, the workflow executor, the agent loop, and durable execution on Inngest.
06	Model provider	The LLM vendors behind the agent loop, model choice, and token cost.
07	Capabilities	The business actions a user can request: search, enrich, signals, the CRM question, and the definition builder.
08	Tools and integrations	The tool registry and the provider jobs that reach third-party data vendors.
09	Context and memory	Scoped context packing, memory, and the platform corpus.
10	Intelligence and product state	Where discovery work lands: global intelligence, then organization prospects, then CRM.
11	Governance	Policy at admission, at every tool call and as cost accrues, plus approvals, budgets and idempotency.
12	Observability and cost	The run row, the spans, the cost meter, and the live events.
13	Storage	Postgres on Supabase, pgvector, and Redis.

Governance and observability apply at every stage, so the diagram shows them as bars with no arrows.

agents

architecture

overview

agentic-infra

inngest

runs

policy

Agentic infra

Agentic infra

The path of one request

Modules

Contents