Flexx Agent
A unified chat assistant that already has hands (Remote), models (Router), and durable context (Memory) wired together. Bring your own models and your own tools.
What Flexx Agent is
Flexx Agent is the assistant that uses all three pillars. Where you would otherwise configure Remote, Router, and Memory separately into a client, the Agent already has them connected:
- Hands — when a machine is connected, the agent can run commands and touch files on it.
- Models — every model you register in the Router is one you can chat with.
- Memory — it remembers you across sessions, machines, and clients.
Open it at app.flexx.dev/agent.
What it needs
| Piece | Required? | Notes |
|---|---|---|
| A registered model | Yes | Add a provider key in Models; no model = nothing to chat with |
| A connected machine | No | Without one the agent can still chat — it just can't act on a machine |
| Memory | Already on | Nothing to configure |
Chat works without a machine
A connected machine is what gives the agent hands. You can start with models and memory only, and connect a machine — by installing the CLI — whenever you want the agent to run commands and edit files for real.
Getting started
The first-run flow walks you through it:
- Connect a model provider — paste an Anthropic, OpenAI, OpenRouter, Groq, or any OpenAI-compatible key. See Bring your own models.
- Pick your default model — used for new sessions.
- Connect your machine (optional) — install the flexx CLI on a machine and connect it, then choose it as your default. The page updates live as the machine comes online.
- Start chatting — memory is already on.
How it works
- Sessions — each conversation is a session. Model calls stream in real time, and tool calls and reasoning are shown inline (and can be collapsed).
- Tools run on your machine — shell, files, processes, and screenshots execute on the machine selected for the session, not on flexx's servers. What tools are available depends on that machine's capabilities.
- Connectors — add extra tools such as browsing or GitHub from the Connectors page.
- Approvals — actions that need consent surface as approval requests you can accept or reject; sessions can be stopped at any time.
- Attachments — you can send images (screenshots, diagrams) alongside a message.
- Runs and activity — past sessions and their work are available in the dashboard's activity view.
Models and budgets
Every model call goes through your Router, so it is subject to the same keys, allowlists, and spend caps you set there. You can change the model per session from the picker, or set a default for the whole org.
Flexx Agent vs. your own MCP client
They use the same pillars:
- Use your own client (Cursor, Claude, ChatGPT, …) when you want agentic coding or chat inside a tool you already live in. See MCP configuration.
- Use Flexx Agent when you want one assistant that already has hands, models, and memory connected — no per-piece setup.
Either way, Memory is shared: what one learns, the others can recall.
Bring Your Own Models (Router)
Add your own model providers — Anthropic, OpenAI, OpenRouter, Groq, or any OpenAI-compatible endpoint — and use them in Flexx Agent or from your own code through one router.
System Service
Run flexx as a persistent system service on Linux (systemd) or macOS (launchd). Auto-start on boot.