The Composer (Input Area)

In the Composer, you configure your Run before sending the message.

1. Selectors (Pickers)

Before sending a message, you choose who should respond:

• Agent or Provider: Select a specialized agent or a direct AI provider (e.g., OpenAI).
• Task or Chain: Select a specific task or a multi-stage workflow (Chain).
• Preselection: Ontheia remembers your last selection for new chats. You can permanently change this default in your user settings.
• Saved per chat: Within an existing chat, your selection is saved for that specific chat and persists across chat switches and page reloads. Switching to an agent automatically adopts that agent’s administrator-configured tool authorization (see section 3) for this chat — overwriting any authorization you set manually before. Switching to a provider instead leaves the current tool authorization unchanged.

2. Text Input

The input field supports multi-line text.

• Send: Press Enter to send or Shift + Enter for a line break.
• Cancel: While a Run is in progress, the send button turns into a Stop button, with which you can immediately cancel the generation.

3. Tool Authorization (Security)

If an agent wants to perform an action (e.g., send an email or read from the calendar), an Authorization Banner appears directly above the input field, depending on the configuration.

• Details: You can see exactly which tool the agent wants to use and which data (arguments) it is transmitting.
• Decision:
  • Allow Once: The agent may perform this single action and then continue.
  • Decline: The action is blocked. The agent receives an error message and usually tries to solve the task without this tool or aborts.
  • Always Allow: (If available) Unlocks authorization for this tool permanently for this Run.
• Queue: If an agent plans several actions simultaneously (e.g., in a Chain or through delegation), you will see a counter display (e.g., +2), signaling how many more authorizations are pending.

4. Prompt Templates

Using the template icon in the Composer, you can save and reuse frequently used texts.

• Save: Saves the current input text as a new template with a title.
• Load: Inserts the template text directly into the input area.
• Template Variables: Templates support variables in the form ${variable} or {{variable}}. They are automatically replaced with current values when inserted.

Variable	Example Output
${user_name}	Jane Doe
${user_email}	jane@example.com
${current_date}	Wednesday, March 25, 2026
${current_time}	14:30
${agent_label}	Master Assistant
${chat_id}	UUID of the current chat

Further variables (e.g. ${agent_id}, ${task_id}, ${role}) are also available — see the full list in the Admin documentation under Tasks / Configuration.

5. Token Display (Usage)

After the agent has responded, a small statistic appears right-aligned under the message in the form ↑ 1.2k ↓ 80 — extended with ⚡ 7.7k on cache hits.

• ↑ (Input): Total number of tokens sent to the model for this run — accumulated across all intermediate steps (e.g., multiple tool calls, chain steps, or delegated sub-agents), including any reused cache tokens.
• ↓ (Output): Total number of tokens generated by the agent, also accumulated across all intermediate steps.
• ⚡ (Cache): Shown only when the model reused tokens from the prompt cache. Indicates how many of the input tokens (↑) came from cache — these are billed by the provider at a steep discount (often ~10%), so a higher value means lower cost. The tooltip states the exact share (e.g., “7.7k of 28.8k input tokens served from cache”). Caching kicks in on repeated requests with a stable context; if the ⚡ is missing, it was a cache miss (e.g., first request or changed context).
• Compact notation: Values from 1,000 upward are rounded (e.g., 12.4k, 1.2M).
• Copyable: Like the timestamp, this display can be selected and copied — useful for sharing values for debugging or support.
• Purpose: Helps you estimate token consumption and thus the cost/load of the run — especially for chains and delegation, where multiple model calls are summed up.

6. Context Size (Composer Bar)

On the right side of the bottom composer bar, before the tool authorization icon, Ontheia shows the current context size as a compact number (e.g., 8.9k T).

• Meaning: Number of tokens actually sent as the prompt to the model in the last request (incl. cache tokens) — independent of tokens consumed by delegated sub-agents.
• Hover: A tooltip shows the exact value in tokens.
• Purpose: Gives you a sense of how close you are to the selected model’s context limit — unlike the cumulative token display on the message (see above), which targets the cost/billing of the entire run.