Outborn Agent Platform · Step-by-step guide
Chat and tool use
Work with an agent, inspect its evidence and tool activity, and review proposed changes before execution.
Before you begin
- Select an agent and Space you can access.
Follow the workflow
- 01
Ask a focused question
Name the outcome and relevant time period or source. Keep secrets out of the conversation.
- 02
Inspect what ran
Review the tool execution, source context, and evidence presented with the answer.
- 03
Handle an action proposal
If a connected write needs approval, inspect its target and effect in the configured approval flow before deciding.
- 04
Record feedback
Mark an incorrect or incomplete response for follow-up and retain the conversation context needed to reproduce it.
You can explain how the answer was produced and whether any proposed action executed.
Distinguish missing source access from unavailable tools, model errors, and pending approval.
Applied guidance
Best practices for chat and tool use
Recommended
- Ask a question with a source or time range that can be checked.
- Read tool results and source evidence before acting on a generated answer.
Avoid
- Do not put credentials in prompts or interpret a proposed action as an executed one.
01
Start a conversation
Select an agent in the intended Space and ask a focused question. Give the agent enough context to identify the task without placing credentials in the prompt.
02
Read tool activity
When an agent uses an available tool, inspect the tool result and source context alongside its answer. Tool availability follows the current user, agent, Space, connection, and policy.
03
Review actions
A proposed external write may require an approval. Check its target, inputs, expected impact, and reviewer before authorizing it.
04
Give feedback
Conversation feedback and execution metadata are retained as product audit events for operational review.
Related documentation