Evaluations¶
An Evaluation is a scheduled instruction that your assistant can think about and act on automatically. It’s what allows your assistant to check things, make decisions, and use its tools in the background — even when you’re not chatting with it.
Evaluations make the assistant proactive. Instead of waiting for you to ask, it can reason about a task, decide if something has changed, and then use the right tools to respond.
How It Works¶
Each evaluation contains a short instruction, such as:
“Check Bitcoin’s price every hour and alert me if it rises above $70,000.”
At the chosen time, the assistant reads this instruction and decides what needs to happen. It can use any of the Tools it has available — both Native Tools and External Tools — to complete the task.
- Native Tools let the evaluation interact with the app itself, for example by storing information, updating local data, or sending you notifications.
- External Tools connect it to outside systems through MCP, such as checking an API or reading a remote file.
By combining these, evaluations can perform complex background checks that involve both your local environment and connected services.
Typical Uses¶
| Example | What It Does |
|---|---|
| Price Monitor | “Check Bitcoin’s price every hour and notify me if it crosses $70,000.” |
| File Watcher | “Every 10 minutes, look in this folder for new files and summarize them.” |
| System Health Check | “Every morning, verify that all servers are online and alert me if any are down.” |
| Daily Summary | “Each evening, create a summary of today’s conversations and updates.” |
| Local Maintenance | “Once a day, re-ingest local documents or refresh the index.” |
Why It’s Powerful¶
Evaluations aren’t just timers or reminders — they think. Because the model reads and interprets your instruction, it can decide whether something has changed and which tools to use next.
An evaluation can remember what happened before, repeat on a schedule, and combine different tools to keep your system up to date. For example, it might use a native tool to log results locally, then an external one to send them elsewhere.
In Summary¶
- Evaluations let your assistant think and act automatically on a schedule.
- They can use both Native Tools (for actions inside the app) and External Tools (for connected services).
- They can reason about what to do, remember past runs, and keep your assistant informed and responsive.
Evaluations are how your assistant stays aware, active, and useful — even when you’re not watching.