AutoGPT icon indicating copy to clipboard operation
AutoGPT copied to clipboard

Logs, Watchdog, Monitored Continuous Mode [auto-PAUSE]

Open fire17 opened this issue 2 years ago • 5 comments

Hi there! Looked around the open issues but haven't seen this suggestion yet

Firstly, I'd like to know that whatever is running is always logged (maybe that's addressed somewhere)

Secondly , I'd like a "Cautious" Monitored Continuous Mode Where, a GPT 3.5 or other model, can see the live logs, and will trigger a PAUSE on the system for the following reasons:

  • Exceeding Token Costs
  • Runaway ML scenerios
  • Potentially Harmful/Immoral procedures

I'd like to have some settings for time delays between executions or inside parts of the loop. And ultimately the ability to Pause a run, Save it, (potentially edit its state while not running), and Resuming the run finally. The pause can be triggered by the monitoring system, and resuming should only happen with user authorization.

Once this is a thing alerts would be the next step

fire17 avatar Apr 04 '23 06:04 fire17

Added simple persistent logging in https://github.com/Torantulino/Auto-GPT/pull/124 via sqlite3 db. Allows for saving and overwriting sections of the "persistent memory".

EricFedrowisch avatar Apr 04 '23 14:04 EricFedrowisch

the ability to Pause a run

... Save it, (potentially edit its state while not running), and Resuming the run finally. The pause can be triggered by the monitoring system

What do you mean By "edit its state while not running"? Do you mean editing previous states? If you just want the ability to pause a run, input human feedback, and continue the run.... then this pending PR #2069 meets the request above.

Either way, the above PR at least introduces a "pause executions in a continuous loop" functionality, which can be used as the pause.

"Cautious" Monitored Continuous Mode

I'd like to expand on that, if the model pauses itself during a loop, I would like it to send me an text or an email to let me know I'm needed. I've started adding Twilio SMS functionality to another branch of mine, so I can have it text me when paused. If that is of interest to you, please let me know and I can prioritize that further. NOTE: Adding SMS/Call/Email providers is a slippery slope to spam enablement, but that's out of the scope of this conversation.

adrianegraphene avatar Apr 21 '23 17:04 adrianegraphene

I'd like to have some settings for time delays between executions or inside parts of the loop. And ultimately the ability to Pause a run, Save it, (potentially edit its state while not running), and Resuming the run finally. The pause can be triggered by the monitoring system, and resuming should only happen with user authorization.

This is basically related: introducing dedicated constraints for system-stats in the form of quotas that the system can internally use to weigh its operations: #3466

Boostrix avatar Apr 30 '23 12:04 Boostrix

I’m your first github follower apparently!

This sounds right but it makes me think of systemctl hahah. When is AutoAutoGPT

On Sun, Apr 30, 2023 at 07:04 Boostrix @.***> wrote:

I'd like to have some settings for time delays between executions or inside parts of the loop. And ultimately the ability to Pause a run, Save it, (potentially edit its state while not running), and Resuming the run finally. The pause can be triggered by the monitoring system, and resuming should only happen with user authorization.

This is basically related: introducing dedicated constraints for system-stats in the form of quotas that the system can internally use to weigh its operations: #3466 https://github.com/Significant-Gravitas/Auto-GPT/issues/3466

— Reply to this email directly, view it on GitHub https://github.com/Significant-Gravitas/Auto-GPT/issues/151#issuecomment-1529010181, or unsubscribe https://github.com/notifications/unsubscribe-auth/AD2U6HJEO33XLOD3QUK2VVLXDZINTANCNFSM6AAAAAAWSIOXXA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

-- Sincerely, Jason P. McElhenney

zudsniper avatar Apr 30 '23 12:04 zudsniper

I'd like a "Cautious" Monitored Continuous Mode Where, a GPT 3.5 or other model, can see the live logs, and will trigger a PAUSE on the system

looking at the sub-agent and agent messaging API, that would also seem like the ideal mechanism to monitor a sub-agents logs in order to pause it temporarily or to restart it if needed

Please help take a look at the following PR #3083, hopefully this can then be closed or made more specific ?

Boostrix avatar Apr 30 '23 17:04 Boostrix

This issue was closed automatically because it has been stale for 10 days with no activity.

github-actions[bot] avatar Sep 17 '23 01:09 github-actions[bot]