Precise description for go_back tool

Update

RRemove old QwenModel

Update

Adapt max tokens

Fix

Add repetition warning

Implement replay function

Improve prompt further

Improve prompting for massively improved results

Ste hyperbolic as fallback

Transfer task details to system prompt

Prepare 2-step crosshair

Add last click marker on screenshots

Working export log to dataset

Add end of step screenshot in chat

Working inference

Re-add images in inference. Inference endpoint model still broken context length

Clarify logs

Try logging pointer clicks

Working agent with saving after task'

Repair second task

Switch to agent logging

Update smolagents to log planning step

Remove end of run errors, simplifies app.py by pruning elements

Refactor e2b agent to make system prompt dynamically accept tool descriptions

Simplify agent code

Update no-provider fallback to Qwen VL 32B endpoint

adding terminal

less verbose

a

adding planning to logs

stoping agent when desktop not available

adding api calls with fallback

updates

first running version