e1e170a5fd
fix agent not stopping: consecutive_failures counter was resetting on multi-action error steps
2026-02-08 17:25:39 -05:00
880b636422
Merge branch 'main' into 3895
2026-02-08 11:15:44 -05:00
9e95241bbb
stop agent if browser closed
2026-02-08 11:14:49 -05:00
79501fe2cd
added message compaction for agent
2026-02-05 12:26:11 -08:00
995cb0176b
updated default max step to 500
2026-02-05 11:20:47 -08:00
cec4aea3bc
increased threshold for loop detection
2026-02-04 16:34:42 -08:00
b96b9436fc
fixed agent loop detection
2026-02-03 16:00:08 -08:00
bd4ead5d2e
Action Loop detection
2026-02-03 15:19:59 -08:00
da5b4237ba
fixed lint issue
2026-02-02 23:01:52 -08:00
0dba38ed95
Merge branch 'main' into align-agent-success-with-judge
2026-02-02 22:16:11 -08:00
8565c7d267
updated agent success with simple judge
2026-02-02 22:14:45 -08:00
65405d8a79
Merge branch 'main' into align-agent-success-with-judge
2026-02-02 21:50:02 -08:00
4d5976c427
Merge branch 'main' into improve-agent-planning
2026-02-02 21:45:06 -08:00
dae6dc6d0e
update agent success with judge verdict
2026-02-02 18:34:11 -08:00
023367052c
in flash mode disable planning
2026-02-02 18:28:28 -08:00
4a522a6f1c
dynamic planing exploration
2026-02-02 16:26:50 -08:00
defe37dd28
Improve multiact for agent
2026-02-02 12:50:03 -08:00
5ad365e31b
added basic planing to agents
2026-02-02 12:28:29 -08:00
db832bcb83
added extraction schema
2026-02-01 20:26:50 -08:00
9b0d2f685f
fixed prompt voilation of under 10 step instruction
2026-01-30 16:57:00 -08:00
65c59be263
add step budget awareness to prompt
2026-01-30 11:09:06 -08:00
3c927b4f9b
Merge branch 'main' into increase-step-timeout-180
2026-01-29 02:06:24 -08:00
7d66592a0c
Increase default step_timeout from 120s to 180s
...
Production data analysis (24h) shows:
- 36.5% of consecutive failures are AsyncIO CancelledError in browser state retrieval
- Complex pages can take 30-60s for DOM extraction + 30-60s for LLM call
- 120s was too tight for complex pages with slow network
180s (3 min) gives ~50% more buffer while still catching stuck steps.
This is the per-step timeout, not the total task timeout.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com >
2026-01-29 01:07:08 -08:00
40336731d5
Remove specific timeout for 'browser-use' model
2026-01-29 00:39:05 -08:00
3b2b213dd3
Increase LLM timeout to 75s for gemini and browser-use models
...
Based on production data analysis showing:
- gemini-3-pro-preview has 48% timeout rate with median 23s/step
- browser-use-2.0 timeout tasks have 74% slower time per step (10.7s vs 6.0s)
- Most timeouts occur on steps 11-20 where tasks are making progress
The 45s default was too aggressive. 75s provides ~3x the median step time
as buffer while still catching genuinely stuck tasks.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com >
2026-01-29 00:35:56 -08:00
be6873636e
clear agent eventbus, not just browser session's
2026-01-28 20:28:14 -05:00
5a3f8b4d54
Merge branch 'main' into fix/respect-use-vision-in-judge-fix
2026-01-27 18:22:06 -05:00
f85bbb49c8
pass model_name to SystemPrompt
2026-01-23 16:33:28 -05:00
147db04c1c
Merge branch 'main' into add-ax-name-fallback-rerun
2026-01-15 11:03:27 -08:00
9339935c1f
added check for history item state
2026-01-15 10:50:07 -08:00
9c042d7d98
make event bus timeout configurable using env
2026-01-15 10:44:44 -08:00
97d71b84fa
added a fix for multi menu dropdown
2026-01-15 10:17:59 -08:00
d73bca47e8
fix: respect use_vision=False in judge to prevent screenshot leakage ( #3726 )
...
When use_vision is set to False, screenshots should not be sent to the
judge LLM. This is a privacy concern as websites may contain sensitive
data that users explicitly do not want shared with LLMs.
Changes:
- Add use_vision parameter to construct_judge_messages()
- Only encode and include screenshots when use_vision is not False
- Pass self.settings.use_vision from _judge_trace() to construct_judge_messages()
- Add comprehensive tests for all use_vision modes
Fixes #3726
2026-01-08 00:44:36 +05:30
2c06eabf5f
Added ax-name fallback for history rerun.
2026-01-06 12:07:10 -08:00
319d0acd45
fixed the edge case for handling max index
2025-12-23 19:29:43 -08:00
2aa23f01dc
fixed backend_node_id based element counting.
2025-12-23 17:25:38 -08:00
0da9457f6b
feat: added exp backoff retry and minimum element load
2025-12-23 17:02:28 -08:00
bb13679f85
feat: remove redundant retry step from history
2025-12-23 16:25:52 -08:00
07b7a6d8c0
fix: include shadow DOM form elements in selector_map even without snapshot data
2025-12-19 18:32:40 -08:00
cf0894ff8d
add support for external skillservice
2025-12-18 19:10:47 -08:00
b9c8346499
fix: change max_step_interval to 45s
2025-12-18 16:45:21 -08:00
afa5b71747
feat: added Unique attribute match and better error logging for rerun
2025-12-17 11:20:46 -08:00
1779e9717d
improve hashing of elements using cascading hash
2025-12-17 00:55:27 -08:00
05b41f64a9
Dont skip failrues by default
2025-12-16 20:04:28 -08:00
e9c047be92
include ax_name for rerun history element hash
2025-12-16 17:05:48 -08:00
dc0db88215
kv caching for BU agents
2025-12-15 18:23:22 -08:00
dd6aa26301
Merge branch 'main' into mert/rerunhistory
2025-12-14 13:36:12 -08:00
a0996ba39e
max wait between execution
2025-12-14 13:35:27 -08:00
cba65772f8
Fix XAI vision check to only disable for grok-3 and grok-code models
...
Previously all grok models had vision disabled. Only grok-3 variants
(grok-3, grok-3-mini, grok-3-fast) and grok-code models don't support
vision. grok-2 and grok-4 models now correctly retain vision support.
2025-12-13 12:40:21 +01:00
20247d5c56
make coordinates optionally available
2025-12-12 14:26:13 -08:00
Laith Weinberger