> For the complete documentation index, see [llms.txt](https://docs.algenta.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.algenta.ai/trust-and-security/architecture.md).

# Architecture and trust boundaries

Where the engine, your model, and your MCP clients run relative to each other — and the one boundary every deployment mode shares.

This is the diagram version of the [shared-responsibility table](/trust-and-security/overview.md#shared-responsibility-by-deployment-mode): the same facts, drawn as a data-flow so the boundary between "yours" and "Algenta's" is a line on a picture instead of a row in a table.

## Self-hosted, the default and recommended path

Every integration, example, and default in this documentation set points at a deployment you run. There is no step where consuming Algenta requires sending data to Algenta-operated infrastructure.

```mermaid
flowchart TB
    subgraph customer["Your infrastructure — laptop, datacenter, or your own cloud account"]
        direction TB
        subgraph engine_box["Algenta engine (self-hosted)"]
            api["api — FastAPI, /v1/*"]
            worker["engine — async worker + compiled Mojo runtime"]
            db[("Postgres — decisions, audit log, policy snapshots")]
            store[("Redis + S3-compatible object storage")]
            api --- worker
            api --- db
            worker --- store
        end

        subgraph model_box["Your model / LLM"]
            direction TB
            localmodel["Local inference (Qwen3-family, Metal decode) inside the engine"]
            byok["BYOK provider (your own OpenAI/Anthropic/etc. credential)"]
        end

        mcp_clients["MCP clients — Cursor, Claude, agent frameworks, your own app"]

        idp["Your identity provider — Entra / Okta / Auth0 / Keycloak / API keys"]

        mcp_clients -- "MCP: stdio or /mcp (Streamable HTTP)" --> api
        api -- "chat/embeddings requests" --> model_box
        idp -- "bearer token / license JWT" --> api
    end

    egress{{"Egress guard — 5 classes, SSRF hard-blocks"}}
    api -. "outbound calls, classified and gated per deployment mode" .-> egress
    egress -. "connector traffic you configured" .-> external["Your configured connectors (databases, object storage, code repos, ...)"]

    style customer fill:#00000000,stroke-width:2px
    style engine_box fill:#00000000
    style model_box fill:#00000000
```

Everything inside the outer box is infrastructure you control. Nothing crosses out of it except:

* traffic to destinations *you* configured (a connector, a BYOK provider endpoint) — gated by the [egress classification](/trust-and-security/data-boundaries.md), never blanket-allowed;
* software distribution (pulling the image or package itself) and commercial/license-management metadata, neither of which is your runtime decision data — see the [subprocessor statement](/trust-and-security/overview.md#subprocessor-statement).

## Where MCP fits

MCP is the primary, mature integration surface (see [Protocol compatibility](/trust-and-security/protocol-compatibility.md)): an MCP client — a coding assistant, an agent framework, a gateway like LiteLLM's MCP Gateway — talks to your self-hosted `/mcp` endpoint, authenticated the same way any other API caller is (see [Identity and access](/trust-and-security/identity-and-access.md)). The governed-execution boundary (policy gates, approvals, receipts) is enforced inside the engine on every path that reaches it, MCP included — it is not a property of the transport, so it applies identically whether the caller is an MCP host, a direct HTTP client, or the Responses/Chat Completions surface.

## What this diagram does not claim

* It does not depict a highly-available, active-active engine. The `engine` component is not active-active today, and the [Helm chart](/trust-and-security/capability-status.md) makes that explicit rather than hiding it: `engine.replicaCount` is hardcoded to `1` and enforced at the schema and template level, not merely defaulted.
* It does not depict Foundry, Toolbox, or any hosted-agent-platform integration as running *inside* this boundary — those are separate, ecosystem-specific deployments that call into a self-hosted engine from outside it, documented per-ecosystem in the framework integrations themselves.
* SaaS and hybrid-provider deployment modes are real but are the exception this documentation set does not default to — see the [shared-responsibility table](/trust-and-security/overview.md#shared-responsibility-by-deployment-mode) for how the boundary moves in those modes.

## Next steps

* [Data boundaries](/trust-and-security/data-boundaries.md)
* [Identity and access](/trust-and-security/identity-and-access.md)
* [Capability status](/trust-and-security/capability-status.md)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.algenta.ai/trust-and-security/architecture.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.