Getting Started

Sign in

Agentic Chat runs in the browser at agentic.imitation.works. You sign in with email/password, Google, or GitHub.

Your first chat

1. Select New Chat in the top left (or press Cmd+K).
2. Pick a model on the right (see Settings).
3. Type your message and send it with Enter (Shift+Enter for a new line).

Which mode do you need?

Auto mode, recommended. Simply describe what you need: the app picks the right path itself: a plain conversation, web research with sources, or data or contract analysis. You don't have to switch anything manually.

Multi-step tasks work in one go too. Phrase them in order:

• "Research the current funding programs for startups in Vienna, summarize the most important ones, and write a short overview from that."
• "Analyze the attached table and turn it into a report with the key figures."

The result appears as a normal message in your chat and stays available after a reload or a chat switch.

You can also set the mode explicitly in the composer:

• Chat for a direct conversation without web research or file analysis: brainstorming, drafting, rewriting, or shortening text; an explanation or summary of a text you provide yourself; quick questions where up-to-date sources don't matter. It is the fastest option and uses no credits with the standard models.
• Research for explicitly sourced web research: current events, figures, or facts that should be verifiable; market, competitor, or topic research with citations. This forces research even when a question would otherwise be handled as a plain conversation. The answer includes citations inline in the text and a source list at the end. See Research Agent.
• Agent runs a custom agent you created.

Projects

Chats, research documents, files, and artifacts can be organized into projects. A project bundles three kinds of context:

• Project description (optional): applies automatically to every chat response within the project, without you re-entering it.
• Files in the project: files assigned to a project can be attached as context for chats (see Working with Results).
• Research documents in the project: results of research runs, stored as durable, reusable knowledge within the project.

This is how the context for a chat response is built:

project description
+ attached files
+ selected research documents
= context for the model

Agents and research do not use this attached context directly. They produce research documents that you can then attach as context in a chat.

Attaching files and research

The plus button in the composer is the entry point for attaching extra context to your next message. On a phone it opens a menu with both options; in the browser it opens the research document picker, while files are added via the paperclip (or Cmd+U).

• Attach files: include uploaded or new files as context (see Working with Results).
• Attach research documents: select stored research documents from the project.
• Attach images: have images analyzed directly (see Vision).

Each message allows at most 5 attachments (files and research documents combined), of which at most 3 research documents. Example: 5 files and no document, or 2 files and 3 documents.

Reusing results

Every answer can be reused:

• Save as a file (Markdown, PDF, Word, Excel) via the save icon on the answer.
• Send by email: a finished agent result straight to you or another address.
• In projects, results become reusable knowledge. Later just ask "Summarize what we've found so far" or "What are the open items?".
• Attach research documents to bring earlier results into a new chat as context.

Tips for better results

• Be specific: state the goal, scope, and desired format, such as "as a table" or "in three sentences".
• One clear task per message in chat; for multi-step tasks, name the steps in order.
• Attach only relevant files, not whole archives; focused context yields better answers.
• Use projects to keep related chats, files, and results together.
• Check sources: the AI can make mistakes, so always double-check important information.
• On your phone: press and hold a message to open its actions (copy, regenerate, delete). When the input field is empty the microphone appears; as soon as you type, the send arrow.

Interface language

The interface is available in German, English, French, Serbian, and Italian. You can change the language in the settings; it affects only the interface, not the selected model.

Subscription & credits

An active subscription is required for agents, Voice, and advanced models. Details under Credits & Subscriptions.