Agents

Every Aderra chat already works like an agent: it plans, uses tools in several steps and carries the job through to the end. An agent you build is a saved version of that with your own instructions. Make one when you keep giving the same setup: “you are my copy editor, British spelling, never change quotes”, or “review my SQL and suggest indexes”.

Your agents are private. They can’t be shared with other people, and there is no public agent catalogue.

Create an agent

1. In the left rail, open Agent Builder.
2. Fill in the form:
   • Agent name and, if you like, a short description of what it does. Click the round image to upload an avatar.
   • Model (required). See Choosing the model below. Category is already filled in, and you can leave it as it is.
   • Instructions (required). This is what the agent is told before every conversation. Be specific about its job, its tone and the format you want back.
   • Tools, if the agent needs them. See Tools for an agent.
   • File Context: files whose full text goes into the agent’s instructions, so it has them on every message. It suits a few key documents, like a style guide or a product sheet. You can add files only after the agent has been created. For a document you only need in one conversation, attach it in the chat instead.
3. Click Create.

To change an agent later, pick it from the list at the top of Agent Builder, edit it, and click Save. Version lists the saved versions of the agent. You can restore an earlier one, which replaces the current setup after you confirm. The buttons next to Save duplicate or delete the agent.

Variables in instructions

The Variables menu above the Instructions box adds values that are filled in each time you send a message: Current Date, Current Date & Time, UTC ISO Datetime and Current User (your display name).

Chat with your agent

In Agent Builder, pick the agent from the list and click Select. A new chat opens with it. The chat is saved in your history like any other, so you can come back to it from Chats.

An agent’s chat has none of the controls from the composer’s badge row: no Unfiltered switch, no Thinking menu, no Artifacts switch and no Image model menu. The agent always uses the model and tools you gave it.

Choosing the model

Click the Model field, choose the provider Aderra, then a model:

Model	Same as in chat
glm-5.3-flash	GLM-5.3 Flash, the standard model
glm-5.3-flash-uncensored	GLM-5.3 Flash Unfiltered

Use one of these two. The list shows other model names as well. They’re billed at their own rates, and some cost several times more per message. To make an uncensored agent, pick glm-5.3-flash-uncensored.

Tools for an agent

A saved agent only gets the tools you add. Nothing is included by default, not even images. Under Tools, click Add and choose from:

• Artifacts: lets the agent build pages, charts and apps that run beside the chat. See Build apps and pages.
• Ask User: lets the agent pause mid-task to ask you a question, with choices to pick from or a box to type in. Answer and click Submit, or click Skip.
• aderra-image: Aderra’s image and video tools, including the uncensored ones and design mockups. Agent chats have no Image model menu, so images are always made at Best quality (300 credits each). Video needs Pro or Max. See Images and video.
• Your connectors: servers you added on the Connectors page. Each of their tools asks before it runs. See Connectors.

Aderra’s own image and video tools follow the same rules as in regular chat: at most 4 images and 1 video per reply, a question before every video, and a question before them once a connector tool has run in the reply. See Tools in chat.

Tools that need your own key

The Add list also shows tools from other services: Calculator, Google, Tavily Search, Traversaal, Wolfram, OpenWeather, DALL-E-3, OpenAI Image Tools, Gemini Image Tools, Flux, Stable Diffusion and Azure AI Search. All except Calculator ask for your own API key for that service, and that service bills you, not Aderra. The limits, questions and content rules on this page and on Tools in chat don’t apply to them. Aderra doesn’t ask before they run, and they don’t count toward the image cap.

Skills

A skill is a set of instructions you write once and reuse, like your brand guidelines or a review checklist. Open Skills in the rail and click Create Skill. Then either:

• choose Write skill instructions and give it a name (lowercase letters, digits and dashes, like brand-guidelines), a description of when it applies, and the instructions; or
• choose Upload a skill and pick a .md file, or a .zip or .skill file that contains a SKILL.md.

Write the description with care: it’s what the agent reads to decide when the skill fits. Each skill has an Available to agent switch. When it’s off, agents don’t use that skill in new messages.

In the builder, the Skills section under Tools decides which skills the agent may use: Off, All, or Selected (then click Add skill to choose them). The agent can pick a skill by itself when it fits. In any chat you can also type $ at the start of a message and choose a skill for that message.

Advanced

Click Advanced at the bottom of the builder.

• Max Agent Steps: how many steps the agent may take in one reply. A step is one call to the model or one round of tool use. A plain answer is one step, and one round of tool use is usually three: the request, the tool, the follow-up. Leave the field empty for the default of 40. You can set a lower number, but not a higher one. If a reply runs out of steps, you’ll see Keep going and Answer now.
• Multi-agent orchestration lets this agent work with your other agents:
  • Subagents: turn on Enable subagents and the agent can hand a focused task to a helper. Allow self-spawn lets it start a fresh copy of itself. Additional subagents picks other agents of yours it may use. Each helper works in its own context with its own step budget, and only a short result comes back.
  • Handoffs: passes the conversation to a specialist agent. Add up to 10 with Add handoff agent. A Handoff description helps the agent choose the right one.
  • Chain: runs a fixed sequence of up to 10 agents, each one reading the output of the ones before.

Each model step is billed, subagents’ steps included, so multi-agent setups cost more per reply.

On the Free plan

Agents work on Free, with these limits:

• 10 messages a day, the same allowance as the rest of the app. See Plans and credits.
• No artifacts. If you ask a Free agent to build something, it tells you that needs a paid plan and links to the Plan page.
• No Ask User. You can add it, but on Free the agent can’t use it.
• No new connectors. Adding servers needs a paid plan. See Connectors.
• No images without a credit pack, the same as in regular chat.