Using Agent-to-Agent

Practical examples of how your AI tools can work with Chalie.

The talk_to_chalie tool

Every interaction goes through a single tool with four parameters:

Parameter Required Description
message Yes What you want to say or ask Chalie
agent_name Yes Your tool’s identity (e.g. “Claude Code”, “CI Bot”)
project_or_task_name Yes The project or task context
loop_in_human No If true, Chalie relays the exchange to the user in its own words after responding. Default: false

When to loop in the human

Set loop_in_human: true when the exchange involves something the user should know about — decisions being made, actions taken on their behalf, or information that affects their day. Keep it false for background lookups and routine coordination.

Loop in: “Tell the user the deploy failed and their 3pm meeting conflicts with the rollback window.”

Don’t loop in: “What timezone is the user in?” or “What’s the project name for the billing service?”

Practical examples

Development workflow

Your coding assistant can pull context from Chalie to make better decisions:

  • “Does the user have any meetings in the next 2 hours? I want to know if I should schedule a long build.”
  • “What’s the current status of the billing service migration? Check the user’s documents and memory.”
  • “Remind the user to review the PR I just opened — it’s for the auth refactor on the payments repo.”

CI/CD integration

A CI bot can keep Chalie informed about deployments:

  • “The staging deploy for payments-api v2.3.1 just completed successfully. All 847 tests passed.”
  • “Build #4521 failed on the main branch. The error is in the auth middleware — test_token_refresh is timing out.”
  • “The release-branch build finished: 52 passed, 3 failed.”

Cross-tool coordination

Multiple agents can coordinate through Chalie as a shared context layer:

  • Research agent: “Store these findings about competitor pricing in the user’s documents — they asked for this yesterday.”
  • Monitoring agent: “The home office temperature has been above 28°C for an hour. Let the user know.”
  • Calendar agent: “The user has back-to-back meetings from 2pm to 5pm. Don’t send non-urgent notifications during that window.”

Personal assistant tasks

Agents can delegate personal tasks to Chalie:

  • “Add milk, bread, and eggs to the shopping list.”
  • “What’s the weather forecast for this weekend?”
  • “Check if there are any unread emails from the accountant.”

What Chalie can access

When your agent talks to Chalie, it has access to a broad set of capabilities — read access to most data sources, and write access to a safe subset. The defaults are deliberately conservative: an external agent can read and organise, but actions like sending email, controlling smart home devices, or running web searches are off by default and require you to enable them in the Policy Manager.

Allowed by default:

  • Memory — preferences, past decisions, personal context
  • Calendar — reading events and checking for conflicts (creating/editing events is off by default)
  • Email — reading inbox and searching threads (sending is off by default)
  • Contacts — looking up people and contact details
  • Documents — creating, viewing, and searching documents
  • Lists — adding to, viewing, and managing shopping lists and to-dos
  • Weather — forecasts and conditions
  • Smart home — reading device state and listing automations (controlling devices is off by default)
  • Scheduling — reading and searching schedules (creating reminders is off by default)

Tips

  • Keep agent_name consistent across calls so Chalie can track each agent’s conversation thread separately
  • Use project_or_task_name to scope the context — Chalie uses it to organise and recall exchanges
  • Chalie maintains conversation history per agent, so follow-up turns work naturally without re-sending context
  • If Chalie doesn’t have the answer, it’ll say so rather than guess