Changelog

Chatting with an agent is easier to follow. Each turn now folds its thinking, tool calls and generated files into one activity timeline, so the answer stays readable.

Answering the agent is faster too. Pick an option and a single-choice question is sent right away. If the agent is waiting on a question, you can just type a message instead.

Reading the docs on a large monitor is more comfortable. On screens wider than 1600px, the sidebar, the page and its outline now sit together in a centered frame instead of drifting to the edges.

Your agent's files are now a place where you can work too. Open a Markdown or text file in the Files pane and edit it like a document, with autosave, while the agent reads and writes the same file. Code, JSON and HTML files are editable too.

Draft a plan together, fix a paragraph the agent wrote, or keep notes the agent can use in its next run.

You can now use your ChatGPT subscription in Agenta. Click the model selector, then Configure models, then choose ChatGPT subscription.

Automations are easier to trust now. Create one on a schedule or an event, run a test right away, and follow every run in its own session.

When a run needs your input, you continue it in that session like any other chat.

When you run many agents, finding the right conversation matters. The sidebar now groups sessions by agent, lets you filter, pin and reorder them, and shows which ones are running or awaiting your input.

Start a new session from any agent with +, and right-click a session for its actions.

Skills are now shared across your project. Write a skill once, publish it, and install it in any agent.

An agent can pin a skill to one version. When the skill changes, the agent shows that an update is available, and you decide when to apply it.

You can now decide how much an agent can do without stopping to ask. You can select whether to allow everything, always ask for approval, or block them.

These are the default permissions for the agent. You can still configure each tool's specific permission to override the default.

You can now attach secrets to your agent without sharing them in chat. For instance, attach your GitHub token so the agent can use gh.

You can also share the agent with your team. They can use it without ever seeing your credentials.

To attach a secret, open Advanced, then Custom secrets, and attach a saved secret or create a new one. When the agent needs a secret during a task, it asks you: click Configure on the request card.

You often want to point your agent to a file in the workspace. Before, you had to find the path and paste it into chat. Now you can type @ in the chat box, search for the file, and select it.

The reference becomes part of the instruction you send.