How to use Hydra

Everything Hydra does, and where to find it. The same guide is in the app under Help.

Asking

Modes: One menu next to the message box decides how Hydra answers. Chat: Smart (default) picks the best model for each message; Fast is quickest; Deep Think reasons at length, then a second model reviews it; Fusion asks every connected model, checks what they dispute and merges one answer; Code writes and runs code; Compare shows each model side by side. Make: Research (a cited report), Build (a website or app), Image and Video. The menu says how long each takes and suggests one for what you are typing.
A single model: Pick a model yourself from the model menu (Model A, B… each with Max, Pro and Swift); the mode still applies. Fusion and Compare use every connected model.
Actions need your OK: When a model wants to send, post, delete, pay or publish through one of your apps, it stops and shows an approval card. Nothing runs until you press Approve and run — and only with exactly the details shown.
Answer options: Inside the mode menu: answer length (Concise, Balanced, Detailed), Web search (live results with sources), Think (the model reasons step by step first) and Second-model check (an independent model reviews the draft, then it is revised).
✨ Improve: Type a rough message and press ✨ Improve next to the paperclip: a fast model rewrites it into a clear, complete prompt. Press again to undo.
👍 👎: Rate answers (or one model's draft inside a Fusion answer). Smart learns which model you prefer for which kind of task — see Settings → Usage.
Side by side: Compare mode shows every model in columns. Inside a Fusion answer, open Individual answers → Show side by side.
Versions ‹ 1 / 3 ›: Retry, Retry with another model and Edit keep the earlier answer (and everything after it) as a version: the arrows switch between them, each naming its model, and Compare lays them out side by side. Nothing is deleted.
Interactive answers: When it helps, an answer comes with something you can use: a live chart (hover or use the arrow keys to read values, switch bars/lines, show the data as a table, download CSV), a sortable table, a calculator whose results update as you change the numbers, a quiz, flashcards, a checklist that remembers your ticks, a timeline, or buttons to choose how to continue. Turn them off in Settings → Usage.
Side question (/btw): Type /btw and a question, or use the speech-bubble button at the top: it's answered in a small panel using this chat as context, without adding anything to the chat. "Add to chat" keeps it after all.
Branch: Under any message: continue the conversation from that point in a new chat. The original stays exactly as it was.
Outline: In a longer chat, the list button at the top shows every question you asked — pick one to jump straight to it.
Your chats: In the sidebar, each chat's More button (or a right-click) renames it, pins it, moves it into a project, archives it or deletes it. Settings is at the bottom of the sidebar, with Automations and this guide.
Voice conversation: With the message box empty, the round button next to it starts a hands-free conversation: speak, hear the answer as it's written, speak again. Tap the circle to interrupt. Every turn is saved in the chat. Needs Chrome, Edge or Safari.
Keyboard: Ctrl+K (⌘K on a Mac) starts a new chat, Ctrl+Shift+F searches chats, Ctrl+B shows or hides the sidebar, Esc stops an answer; arrow keys move between tabs.

Doing

Mission Control: For bigger jobs (sign in first; sidebar → Automations): describe the objective and a budget. A planning agent proposes the steps; nothing runs until you approve them. Specialist agents — research, reasoning, coding, design, analysis, writing — each do their step on the model best suited to it, a verifier on a different model checks the result, and you get one finished deliverable. You can close the page and come back; cancel any time; costs are shown per step.
Scheduled tasks: Run a prompt hourly, daily or weekly (sidebar → Automations). Add an "Only notify me if…" condition (for example "only if it's going to rain tomorrow") and you'll only hear about it when it's true; every run is kept in the task's history.
Your own knowledge: With memory on, or inside a project, Hydra can search your own saved chats, projects and memories when you refer to earlier work — never anyone else's.
Research: Breaks a question into sub-questions, searches each, and writes a cited report.
Build: Describe a website, app or API; it picks a design direction, creates imagery with Higgsfield, writes every file, loads the site to catch errors, reviews and fixes it (two rounds), shows a live preview and publishes it to a real web address.
Pictures and video: Ask in chat ("draw…", "make a video of…") or pick Image or Video in the mode menu. Attach your own photos to change them ("change the background", "make me a cartoon"), up to 10 at once, or to animate one. Every image has Edit, Animate, Remove background, Upscale and Variations. Videos take a length and a camera move; Cinema clips come with sound, run 4 to 30 seconds and can follow up to 10 reference photos.
Images inside answers: Any model in any mode can create an image mid-answer with Higgsfield (a blog header, a product mock-up) — just ask. Built sites get real hero imagery too.
Analyse data: Attach a CSV/Excel/JSON file: the model writes Python, it runs on your file right in the browser, the output goes back to the model and it explains the results — charts included. Errors are fixed automatically.
Canvas: Open any answer or code block in the Canvas: edit it side by side, select a part and say what to change — only that part is rewritten. Quick actions, undo, download.
Files: Ask for a Word document, Excel workbook (with live formulas), PowerPoint, PDF or CSV and download it, or use "Save as…" under any answer. Upload Word/Excel/PowerPoint/PDF/images and every model reads them; a video becomes a sheet of its frames (and a transcript when it's small). A very long paste becomes an attachment automatically ("Paste as text" undoes it).
Recordings: Attach an audio file (MP3, M4A, WAV, WebM, OGG, FLAC — up to 3.5 MB): it is transcribed and the transcript goes to the model, so you can ask for a summary, action items or a reply. Needs an OpenAI key on the site or in Settings.
Run code: Python and JavaScript in answers have a ▶ Run button — results and charts appear inline; errors get an 'Ask Hydra's assistant to fix' button.
Verify: Fact-checks any answer against the web using a different model than the one that wrote it.
Continue · Retry with…: Carry on a cut-off answer, or regenerate it with a different model.
Select any text: Highlight part of an answer → Explain, Simplify, Go deeper, Fact-check or Quote it.
Maths & diagrams: Formulas display properly, and diagrams written in Mermaid become flowcharts, sequence diagrams and mind maps.

Your setup

API keys: Add keys for any of the eleven providers in Settings → API keys. One is enough; Smart, Fusion and Compare use whatever you've connected.
Connectors: Notion, Linear, GitHub, Slack, Stripe, Google & Microsoft (via Zapier/Composio), 50+ more, or any app that offers an MCP server (Settings → Connectors) or any web API (Settings → Custom APIs). Once connected, every model in every mode can use them — including scheduled tasks and agents.
Agents: A saved persona with its own instructions, knowledge files, model, starter prompts and a Higgsfield-made avatar — like Custom GPTs. Sidebar → Agents. Export one as a file to share it; anyone can import it from New agent and check it before creating it.
Projects: Shared instructions and knowledge files for a group of chats.
Memory & custom instructions: Tell it about yourself and how you like answers (Settings → Memory); it also remembers useful facts as you go, and you can edit them all.
Scheduled tasks: Prompts that run on a timetable while you're away and land in your chat list.
Usage: Totals for this month, and which models Fusion and Compare use, are in Settings → Usage.
Bring your history: Settings → Data imports the conversations.json export from other AI chat apps; you can export everything as one file at any time.
Your own prompts: Settings → Prompts: saved templates that become /commands in the message box.

Shortcuts

/: Type a slash in the message box for prompt templates (/summarize, /fix, /email…) and to switch to Research, Build, Image or Video.
Ctrl+K · Ctrl+Shift+F · Ctrl+B: New chat · search chats (it looks inside chats too and opens them at the match) · toggle sidebar.
Temporary: Top-right toggle: a chat that is never saved and never touches memory.
Esc · Undo: Esc stops an answer; deleting a chat shows an Undo for a few seconds.
Listen: Read any answer aloud — a natural voice when an OpenAI key is connected, the browser's voice otherwise.
Share · Export: Share a public read-only link, or export as Word, PDF or Markdown.

Try it: open Hydra. Your first 10 messages are free, with no account.