Multi-agent chat

Every agent you run, in one conversation.

Aura is the chat layer for your AI agents. Connect any OpenAI-compatible provider, give each agent its own instructions, model, and tools, then talk to them by name — in the same thread.

Get startedSee how it works

Sign in with Google or GitHub. Bring your own model key — nothing is billed through us.

RA

Research Assistant

@research-assistant

active
Summarise the three papers I attached and flag where they disagree.
tool · read_document
RA
All three agree on the sampling bottleneck. They split on cause: two blame the scheduler, one blames the dataset. Want the citations?
Yes — just the scheduler ones.

Everything the chat needs. Nothing it doesn't.

Aura owns the conversation — accounts, agents, history, permissions, streaming. Your runtimes keep doing what they do best: running agents.

Bring your own model
Point Aura at any OpenAI-compatible endpoint — a hosted API, a gateway, or Ollama on your own machine. Your keys, your bill.
Agents with a real identity
Name, handle, avatar, instructions, and a default model per agent. Describe one in a sentence and let Aura draft the persona for you.
Mention anyone with @
Type @ to hand the turn to an agent. The list filters as you type, arrow keys move, and Enter selects — no modal, no context switch.
Tools you define
Describe a function as a JSON Schema, then assign it to the agents that should have it. Switch any tool off per agent without deleting it.
Streamed, not awaited
Replies arrive token by token over a typed RPC channel. No polling, no fake progress bars, no spinner pretending to be a response.
Built for more than one person
Direct threads and group conversations, with message history the platform owns — not the model, and not the runtime.

Running in about three minutes.

There is no model to buy and nothing to install. If you already have an API key, you are most of the way there.

  1. Connect a provider

    Add an OpenAI-compatible key and choose which of its models your agents may use. The key is encrypted before it ever reaches the database.

  2. Create an agent

    Give it a name, a handle, and instructions. Pick a default model, then attach any tools it should be able to call.

  3. Start talking

    Open a thread and type @ to bring an agent in. Switch models mid-conversation whenever you want a second opinion on the same prompt.

Your keys never leave your account.

You are handing Aura a credential that can spend real money. It is treated that way.

Encrypted at rest

Credentials are sealed with AES-256-GCM before they are stored, using a key that only the server holds. Plaintext never leaves the process.

Scoped to your account

Providers belong to a single user. Every read path filters on the caller's id, so one account's key is unreachable from another's request.

Guarded egress

Model discovery refuses loopback, private, and link-local addresses, re-checks every resolved address, and will not follow a redirect.

Questions worth asking.

Including the ones with an honest “not yet”.

Give your agents somewhere to talk.

Sign in, add a key, and have your first agent answering in about three minutes.

Get started

Google or GitHub · bring your own model key