Platform Guide

What is aethera?

The Problem

Right now, if you want to use the best AI models, you're stuck paying for multiple subscriptions. ChatGPT Plus is $20/month. Claude Pro is $20/month. Gemini Advanced is another $20/month. That's $60+ per month for three chat windows that do essentially the same thing - and you're still missing DeepSeek, Llama, Grok, and others.

On top of that, every platform locks you in. Your conversations don't transfer. You can't compare how different models handle the same question. And none of them let you build anything beyond basic chat - no agents, no automation, no workflows.

The Solution

aethera puts every frontier AI model from 10 providers into one interface. Claude, GPT, Gemini, DeepSeek, Llama, Grok, Kimi, Qwen, MiniMax, GLM, Mistral - all accessible from the same chat window. Pick any model for any conversation. Switch mid-conversation if you want a second opinion. Free to use, with no subscription and no monthly fee.

More Than a Chatbot

Most AI platforms are just chat windows. aethera lets you build AI Agents - specialized assistants that don't just talk, they act. An agent has a defined role (researcher, coder, writer), a personality, and access to real tools:

• Web Search - the agent searches the internet for current information
• URL Reader - the agent reads and extracts content from any webpage
• Code Executor - the agent writes and runs Python code in a sandbox

Then go further: chain agents into Teams. A Team is a group of agents working on the same task in sequence. The Researcher finds information, the Analyst processes it, the Writer produces the final report. One prompt from you, three specialized AI steps, polished output - automatically.

Why It's Different

• One platform, every model - stop switching between five different AI apps. Every major model in one place.
• Nothing to subscribe to - no $20/month plans, no seats, no tiers. Every model is open to every account.
• AI that does things - agents don't just chat. They search the web, read articles, run code, and produce real results.
• Multi-agent workflows - Teams let you automate complex tasks that would normally require multiple tools and hours of work.
• Agents that run themselves - put any agent on an hourly, daily or weekly schedule, or set one to watch a page and react when it changes.
• Privacy by default - wallet-based login means no email, no password, no personal data. Your conversations are tied to your wallet, not your identity.
• Web + Telegram - same models, same agents, same conversations on both platforms. Chat from your desktop or your phone.

Quick Start

Connect your wallet → pick a model → start chatting. No signup form, no email verification, no credit card. Takes about 10 seconds.

Getting Started

Connect Your Wallet

aethera uses wallet-based authentication. Your wallet address is your identity - no email, no password, no account form.

aethera works with MetaMask, Rabby or any injected browser wallet, and with WalletConnect from a phone. If you have none of them, install MetaMask from metamask.io.

After Connecting

Your account is created automatically the first time you connect. No confirmation email, no waiting. Your username is derived from your wallet address (shown in truncated form, e.g. 0x1a2b…9f3c). Every model, agent and team is available immediately, with nothing to set up.

Chat

Interface Overview

The left sidebar lists all your conversations. Click any conversation to open it, or start a new one with Ctrl+K or the New Chat button at the top of the sidebar. The main area is where the conversation happens. At the bottom is the input area - this is also where you select your model, agent, and team before sending.

Attachments

Click the paperclip in the chat input to attach images (PNG/JPG/WebP) or documents (PDF, TXT, MD, CSV). Vision-capable models read images directly; documents get parsed into the conversation context. Models without vision will tell you they can't see images - switch to Claude, GPT-5, or Gemini and re-send.

Model, Agent & Team Selectors

Below the message input bar you'll see three selectors separated by dots. From left to right: Model (which AI to use), Agent (optional role and tools), and Team (optional multi-agent pipeline). For a standard chat, just pick a model and leave the other two at their defaults.

Streaming Responses

Responses stream in token by token in real-time - you see the answer being written as the model generates it. This means you don't wait for the full response to finish before reading. You can scroll up to read earlier parts of a long response while it's still generating.

Keyboard Shortcuts

• Ctrl+K - start a new chat from anywhere
• Enter - send your message
• Shift+Enter - insert a new line without sending

Switching Models Mid-Conversation

You can change models at any point. The full conversation history is passed to the new model, so it has all the context from previous messages. Keep in mind that longer conversations cost more regardless of model, because the entire history is sent with each new message.

Organizing Your Chats

The chat sidebar groups conversations into folders with optional color tags. Drag any chat into a folder to organize, drag it back out to ungroup. Pin important chats to the top. Use the search bar to full-text search across every message in every conversation - it finds matches inside replies, not just titles.

Available Models

Every model below is available to every account. Pick one per conversation and switch whenever you want a second opinion.

AI Agents

What Makes an Agent Different

A regular chat sends your message to a model and returns a response. An agent goes further: it has a system prompt that defines its role and constraints, it has access to tools it can use autonomously (like searching the web or running code), and it can take multiple steps to complete a task before replying. Instead of just answering, an agent can research, compute, and verify before giving you a result.

Preset Agents

Available Tools

• web_search - searches the internet and returns relevant results with titles, URLs, and snippets. Useful for current events, finding sources, or anything that needs up-to-date information.
• url_reader - fetches any publicly accessible URL and extracts the full text content. Use this to have an agent read an article, documentation page, or any web page you provide.
• code_executor - runs Python code in an isolated sandbox and returns the output. See the sandbox notes below for what's allowed.

Code Execution Sandbox

When an agent runs Python, the code executes in a locked-down environment. Each call starts with a fresh workspace - nothing carries over between runs.

What works: the Python standard library (math, statistics, json, csv, re, datetime, …), data-science libraries (numpy, pandas, matplotlib, scipy, sympy), writing temporary files, and any pure computation.

What doesn't work (on purpose): no internet access, no access to the rest of the server's filesystem, no persistent state between calls, a hard 10-second timeout per call, and output capped at 2000 characters.

Temperature

Temperature controls how predictable vs. creative the agent's output is. 0.0 is fully deterministic - best for coding, data extraction, or any task where precision matters. 1.0 is maximally varied and creative, better for brainstorming or writing. The default is 0.7. For research or analysis, lower temperatures (0.2-0.4) tend to produce more reliable results.

Creating Custom Agents

Go to Agents → "New Agent". Configure a name and role, a system prompt (personality, constraints, and how the agent should approach tasks), the tools it can use (web_search, url_reader, code_executor), and the model and temperature.

Shared Agents

You can share any custom agent as a frozen snapshot, prompt included, so other people can use what you built. Anyone can open the shared tab, clone a snapshot into their own list, and run it. Cloning makes a private copy, and the original snapshot stays immutable, so the version you cloned never changes underneath you.

Teams

How Sequential Execution Works

A team runs agents one after another in a defined order. When you submit a prompt to a team:

• Step 1 - the first agent receives your original prompt and runs. It may use its tools (search, read URLs, run code) before producing its output.
• Step 2 - the second agent receives both your original prompt and the first agent's full output as its input. It processes and extends the work.
• Step 3+ - each subsequent agent receives all previous outputs, building on everything that came before it.
• Final output - the last agent's response is the team's result. You can also review what each individual agent produced.

Preset Teams

Watching Execution in Real-Time

While a team is running you can watch each agent work live. The interface shows which agent is currently active, what tool calls it's making, and streams its output as it generates. You can expand each agent's step to read its full output before the team finishes.

Custom Teams

Go to Teams → "New Team" to build your own pipeline. A custom team is just an ordered list of agents - minimum 2, maximum 6. The first one sees your prompt; every agent after that also sees all previous agents' outputs, so put the research or data-gathering agents early and the polish / review agents last.

Patterns that tend to work: research → analyze → write; code → review; research → write → critique. Keep teams short - longer chains drift, because each step re-sends all prior outputs to the next model.

Telegram Bot

The Telegram bot is launching soon. Everything below describes how it works, so it is ready to follow the day it goes live.

Finding the Bot

Open the bot from the link on the web app's Settings page. Send /start to initialize - the bot will introduce itself and show available options.

Chatting

You don't need any special command to chat. Just type your message and send it. The bot responds using whatever model you've currently selected. Responses are streamed when the model supports it, and longer responses are split into multiple messages automatically.

Switching Models

Send /menu to open the interactive model selector. This shows every available model. Select one and all future messages in that session will use it. Your selection persists between sessions until you change it again.

Commands

Linking Your Account

Linking connects your Telegram identity to your web wallet account. Once linked, the two share one account - start something on the web and carry it on in Telegram, or the other way round. Conversations made in Telegram also appear in your web history.

• Send /link to the bot - you'll receive a short one-time code.
• Open Settings → Telegram in the web app and paste the code.
• Both accounts are now linked and share one history.
• To disconnect later, use the Unlink button in Settings → Telegram.

Settings

Theme

Toggle between dark and light across the interface. Your choice is persisted against your wallet, so the same theme shows up automatically on any device the moment you connect the same wallet. If you haven't connected a wallet yet, the theme falls back to localStorage in the current browser and syncs to your account the first time you connect.

Default Model

Pick which model opens automatically when you start a new chat. Every model stays available inside each conversation regardless of what you set here. Existing conversations are untouched - only brand-new chats pick up the new default.

Telegram

The Telegram section in Settings is where you link or unlink your account. Full walkthrough (/link code, paste flow, shared history) lives in the Telegram Bot section above.

Your Data

Where it lives

Conversations, agents, teams and settings are stored in Supabase (Postgres), encrypted at rest. Uploaded files (images, PDFs, attachments) go into Supabase Storage, scoped to your wallet address. Secrets like API keys and linked-Telegram tokens are per-account and never exposed client-side.

How long it's kept

There is no auto-purge. Everything stays until you delete it. Delete a conversation and the row plus every message in it is hard-deleted immediately. Delete your account and every conversation, agent, team, file and linked Telegram record is wiped in the same transaction.

What can't be deleted

Training

Your messages are not used to train any model. Each request is forwarded to the model's original provider for inference only, and discarded from our side once the response streams back.

Export or deletion request

No self-serve export button yet. Email us at team@aetheraagents.app with your wallet address and we'll send you a full JSON dump of your data - or wipe everything, if you'd rather have it gone.

API

aethera has a public developer API. Mint a key, point any HTTP client at it, and call the same models and agents the web app uses from your own app, bot, or backend. You write the frontend, aethera does the work behind the key.

Get a key

Open Dashboard → Developer API and click create key. The key is shown once at creation (format sk_aero_...) and only its hash is stored, so copy it immediately. You can mint several keys (one per app); they all share one balance. Revoking a key deletes it and stops it working at once.

Base URL and auth

All endpoints live under /v1 on your aethera domain. Send the key as a Bearer token on every request:

Base URL:  https://aetheraagents.app/v1
Header:    Authorization: Bearer sk_aero_...

Endpoints

Example: chat

curl https://aetheraagents.app/v1/chat \
  -H "Authorization: Bearer sk_aero_..." \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-5","messages":[{"role":"user","content":"Hello"}]}'

Response:

{
  "model": "claude-sonnet-5",
  "content": "Hi! How can I help?",
  "usage": {
    "inputTokens": 9,
    "outputTokens": 7
  }
}

Example: agent

Run a tool-using agent. agent is one of researcher, coder, writer, analyst, critic, summarizer. It uses its tools (web search, URL read, code exec) as needed and returns the final answer.

curl https://aetheraagents.app/v1/agent \
  -H "Authorization: Bearer sk_aero_..." \
  -H "Content-Type: application/json" \
  -d '{"agent":"researcher","message":"Latest Base chain TVL trend","model":"claude-opus-5"}'

Response:

{
  "model": "claude-opus-5",
  "agent": "Researcher",
  "content": "Base TVL has...",
  "usage": { "inputTokens": 4120, "outputTokens": 880 }
}

Example: github

curl https://aetheraagents.app/v1/github \
  -H "Authorization: Bearer sk_aero_..." \
  -H "Content-Type: application/json" \
  -d '{"repo_url":"https://github.com/owner/repo","model":"deepseek-v4-flash"}'

Response: a structured verdict object (verdict, confidence, code quality, AI-slop signals, security, activity) plus the usual usage block.

Example: legitimacy

curl https://aetheraagents.app/v1/legitimacy \
  -H "Authorization: Bearer sk_aero_..." \
  -H "Content-Type: application/json" \
  -d '{"project_url":"https://example.xyz","model":"claude-sonnet-5"}'

Returns a markdown report (the verdict) plus usage.

Example: docs

curl https://aetheraagents.app/v1/docs \
  -H "Authorization: Bearer sk_aero_..." \
  -H "Content-Type: application/json" \
  -d '{"url":"https://docs.example.com","model":"z-ai-glm-5-3-flash"}'

Returns a structured report (verdict, slop score, TL;DR, tech stack, socials) plus usage.

Example: slop scan

curl https://aetheraagents.app/v1/slop \
  -H "Authorization: Bearer sk_aero_..." \
  -H "Content-Type: application/json" \
  -d '{"text":"This revolutionary, game-changing tool is fast, simple, and powerful."}'

Choosing a model

Every endpoint that uses an LLM takes an optional model field. Pass any id from /v1/models to pick it; omit the field and the default model is used. A model id that does not exist returns 400. Pure-analysis endpoints like /v1/slop take no model.

Errors

• 401 missing, malformed, or revoked key.
• 400 bad request (unknown model, missing fields).
• 422 the input could not be processed (for example a bad video URL).

Rate Limits

aethera doesn't cap usage per hour or per day. This applies to every model and every feature - chat, agents, teams, web and Telegram - no exceptions.

Failed or aborted requests are discarded rather than saved. A "failed request" means a network error before the model replied, a provider 5xx/timeout before any tokens arrived, or you clicking Stop before the first token. If the model started replying and then cut off mid-stream, you keep whatever actually arrived.

Troubleshooting

Common things that go sideways and how to recover. If none of this helps, email us at team@aetheraagents.app and we'll take a look.

Wallet won't connect

• aethera only supports MetaMask today. Make sure the MetaMask extension is installed and unlocked before clicking Connect.
• On mobile, open aethera inside the MetaMask mobile app's built-in browser - a normal mobile browser can't talk to the extension.
• If the signature prompt never appears, check that pop-ups aren't blocked.

Chat reply cuts off or errors out mid-stream

• Usually a network blip or a provider rate limit on their end. Press retry or send the same message again.
• If it keeps failing on the same model, try a different model: one provider may be degraded while the rest work fine.
• Very long replies can hit the model's output cap. Ask it to continue or break the task into smaller prompts.

FAQ