x1agent

x1agent

x1agent is a Kubernetes-native platform for running LLM agents in production. It provides a security-first container architecture, a pluggable provider system, and real-time bidirectional communication between agents and clients.

Core ideas

Agents run in isolated pods. Each agent session is a Kubernetes Job with two containers: an agent container (untrusted, runs the LLM) and a sidecar container (trusted, holds credentials, enforces permissions). The agent container receives zero secrets.

Providers are swappable. Authentication, knowledge graphs, file storage, messaging, calendars, email — all pluggable. Providers are standalone services that communicate over NATS. Switch from Google Drive to OneDrive by changing a Helm value.

Security is structural, not aspirational. Credential isolation, permission gates, and trust boundaries are enforced by container boundaries and network policy — not application-level checks that can be bypassed.

Architecture at a glance

graph TB
    subgraph pod["Session Pod"]
        agent["Agent Container"]
        sidecar["Core Sidecar (Rust)"]
    end

    nats["NATS"]
    api["API Server"]
    pg["PostgreSQL"]

    subgraph providers["Provider Services"]
        graphProv["Graph Provider"]
        files["File Provider"]
        msg["Messaging Provider"]
    end

    agent -- "localhost" --> sidecar
    sidecar <--> nats
    nats <--> providers
    api <--> nats
    api --> pg
    api -- "creates Jobs" --> pod

    browser["Browser"] <-- "WebSocket" --> nats
    browser -- "REST" --> api

Next steps

• Quickstart — run x1agent locally in under 5 minutes
• Architecture — understand the pod model, communication paths, and trust boundaries
• Providers — how the plugin system works
• Security — the security model in detail