What would you like to work on?
Bring a question, an idea, or a task. We’ll take it from here.
Message Ægentica
What would you like to work on?
Bring a question, an idea, or a task. We’ll take it from here.
Practical ways to use the capabilities already built into Ægentica, plus what needs setup.
Configuration does not guarantee every workflow has been tested. Inspect live capabilities.
Save instructions, reference context and a preferred model; choose a profile in chat or delegate focused work.
Open Agents, create a profile, then choose Chat. Or ask to list saved agents and delegate a research task.
One operator workspace; profiles do not grant permissions. Delegation uses the compiled researcher and current conversation model. Model execution requires a working key.
Ask focused research and review specialists to help with a task using their own context.
Ask: “Research this idea and have the reviewer identify gaps in the result.”
Specialists use model credits and may require additional tools for external research.
Review a proposed tool action before allowing the agent to continue.
Ask: “Create a short note in the workspace and ask me before writing it.”
Approvals are specific to configured tools. Other tools can run without a prompt.
Start a longer review, continue the conversation, and retrieve its result.
Ask the agent to use background_review on a short text.
Background tools are included; individual workflows should be tested before relying on unattended execution.
Reminders, briefings and recurring jobs that run on their own and arrive as a chat and a notification.
Open Tasks → New task, or ask: “Every weekday at 8, give me a short briefing.”
Turn on notifications in Settings → Notifications on each phone or computer that should be told.
View, save and delete images stored on the server volume.
Open Images. Choose Create to start an image prompt in the main composer.
Signed-in operator access. A working provider key is needed for generation, not browsing. Deletion also removes the image from linked conversations.
Ask the agent to save a stable preference and recall it in a future conversation.
Ask: “Remember that I prefer short answers in Swedish.” Then open Memory to see it, or ask about it in a new conversation.
Memory is stored on your server and shared by the app, Telegram and scheduled tasks. Edit or import it on the Memory page.
Bring images, PDFs and text into a conversation, and use a workspace for generated files.
Attach or paste an image, PDF or text in the main composer.
Main chat accepts four files / 6 MB total; unsent drafts stay on this device. The lightweight sandbox is not a full browser or arbitrary Linux environment; workspace files are not guaranteed to survive a deployment.
Maintain task-specific state separately from long-term preferences.
Ask the agent to use its session counter, then increment it again in the same conversation.
Session state belongs to that conversation; profile memory serves a different purpose.
Ask for an explanation, compare options, or turn a broad question into a focused research plan.
Ask: “Help me compare three approaches to my next project. Ask me about my constraints first.”
External sources depend on the available search and fetch tools. Check important claims against their sources.
Open a recent conversation to return to its context and follow a response already in progress.
Start a conversation, return to New chat, then select the conversation under Recent.
History is saved on your server, so the desktop and the installed app share it.
Add direction while work is running, or stop a response when you want to change course.
While Ægentica is responding, type a correction and send it to steer the same turn, or tap Stop.
Stopping generation does not undo tool actions already completed.
Break work into concrete steps, explore tradeoffs, and ask the agent to clarify missing information.
Ask: “Help me plan a small product launch. Ask me about the audience and deadline.”
You decide which proposed actions to take.
Describe a picture, illustration or logo and get it in the chat.
Ask: “Make a watercolor illustration of a hummingbird, landscape.”
Uses the active model provider and costs a little per image.
Dictate with the microphone in the message box and have replies read aloud.
Tap the microphone, speak, and send. Choose a voice in Settings → Voice.
Uses your device's speech features; availability depends on the browser.
Review pull requests, find why a check failed, and open issues or pull requests in your repositories.
Ask: “Review the latest pull request in my repository and tell me what could break.”
Add a GitHub token in Settings → Connections → GitHub. Anything that changes GitHub asks for your approval first.
Add access to tools and context from services such as Linear, Notion and Sentry.
Open Connections for configured accounts, or Capabilities → Explore for the broader directory.
These accounts are not connected. Each integration needs credentials and permissions; a catalog entry is not an active connection.
Use the same agent through Slack, messaging platforms or a custom application.
Open Capabilities → Explore and filter Channels.
Channels need their own accounts, credentials and routing. The web app is the configured entry point here.
Connect external tool servers or cooperate with another agent over HTTP.
Read the connection guide to choose an authenticated integration.
The included MCP server is restricted to local development. Remote agents are not configured.
Model access is metered, with limits on each session and on the API key of the active provider (AI Gateway or OpenRouter).
Keep initial tests short. Test your provider key in Settings.
Each session stops at $25 of model cost. Set a spending limit on each provider key in the provider’s dashboard.
Inspect a live session, test agent behavior and trace a task before using it for important work.
Open Activity for advanced controls or read the evaluation guide.
Reset and clear change a session’s context. External monitoring services require their own setup.