You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bug]: Web Fetch or Search is impaired in releases with/since desktop app!? #42360
Really appreciate the release of the desktop app for Hermes agent however I've noticed that simple requests requiring a web search take an inordinately long time to complete... just asking the agent what's new in the recent version 0.16 release was running over 14 minutes before I attempted killing the process (which also seems to be a problem as the app wouldn't respond/stop). This has been observed on no less than three occasions... By comparison I could switch to OpenCode and get the correct answer(s) in seconds.
Steps to Reproduce
Ask the agent a question that warrants using the web search/fetch functionality to retrieve online information.
Expected Behavior
Reading an online webpage (especially available markdown files) should very quick; quicker than me manually visiting the site and reading for myself.
Actual Behavior
Tool call runs for minutes, most times without a reasonable end.
One PR addresses #42360 by adding timeout wrappers for web search/extract and parallelizing Firecrawl extraction. Its diff targets plausible causes of prolonged calls, but the synchronous search timeout implementation still waits for the worker when leaving the ThreadPoolExecutor context, so it does not reliably bound wall-clock latency.
Merge #42543 after correcting executor shutdown/cancellation semantics so search timeouts actually return within their stated bounds and adding regression coverage for a permanently blocked provider; retain its Firecrawl extraction parallelization and backend extraction ceilings. There are no competing PRs to close as duplicates.
Complex graph
flowchart LR
classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
classDef best stroke-width:3px,stroke:#b45309
classDef target stroke-width:3px,stroke:#4338ca
I42360(["issue #42360 (open)"])
P42543["PR #42543 (closed)"]
P42543 -->|best fix| I42360
class I42360 open
class I42360 target
class P42543 closed
class P42543 best
click I42360 "https://github.com/NousResearch/hermes-agent/issues/42360"
click P42543 "https://github.com/NousResearch/hermes-agent/pull/42543"
Loading
Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed or no verify verdict yet (state tag in the node label).
Cross-PR triage: Reviewed 1 pull request and 1 issue in this complex. Each diff was read against this issue; Assessment working set: 21 kB of PR diffs, 4 kB of issue/PR text, <1 kB of discussion (2 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.
Search and extract are now wall-clock bounded at current main: DuckDuckGo search runs in a disposable process under a 30s deadline (489b85e), web_extract dispatch is capped by web.extract_timeout via asyncio.wait_for (tools/web_tools_extract.py), and every shipped provider sets an HTTP timeout (15–90s). If you still see a web call run for minutes on a build that includes those commits, please reopen with the debug report.
Activity
GottZ commentedon Jul 29, 2026
Summary
One PR addresses #42360 by adding timeout wrappers for web search/extract and parallelizing Firecrawl extraction. Its diff targets plausible causes of prolonged calls, but the synchronous search timeout implementation still waits for the worker when leaving the ThreadPoolExecutor context, so it does not reliably bound wall-clock latency.
Related pull requests
best fix— (+249/-127) — partial fix: The diff parallelizes Firecrawl URL scraping and adds provider-level extraction ceilings, directly reducing serialized multi-URL latency. Despite the APPROVED review on fix(web): cap search/extract timeouts and parallelize firecrawl scrap… #42543, both synchronous search wrappers usewith ThreadPoolExecutor(...); afterfuture.result(timeout=...)times out, context-manager shutdown waits for the still-running worker, defeating the claimed hard cap, while the desktop stop/unresponsiveness path is not addressed; although closed, fix(web): cap search/extract timeouts and parallelize firecrawl scrap… #42543 remains relevant as the only implementation directly linked to [Bug]: Web Fetch or Search is impaired in releases with/since desktop app!? #42360 and the best available fix basis.Suggested consolidation
Merge #42543 after correcting executor shutdown/cancellation semantics so search timeouts actually return within their stated bounds and adding regression coverage for a permanently blocked provider; retain its Firecrawl extraction parallelization and backend extraction ceilings. There are no competing PRs to close as duplicates.
Complex graph
flowchart LR classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937 classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151 classDef best stroke-width:3px,stroke:#b45309 classDef target stroke-width:3px,stroke:#4338ca I42360(["issue #42360 (open)"]) P42543["PR #42543 (closed)"] P42543 -->|best fix| I42360 class I42360 open class I42360 target class P42543 closed class P42543 best click I42360 "https://github.com/NousResearch/hermes-agent/issues/42360" click P42543 "https://github.com/NousResearch/hermes-agent/pull/42543"Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed or no verify verdict yet (state tag in the node label).
Cross-PR triage: Reviewed 1 pull request and 1 issue in this complex. Each diff was read against this issue; Assessment working set: 21 kB of PR diffs, 4 kB of issue/PR text, <1 kB of discussion (2 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.
OutThisLife commentedon Sep 27, 2026
Search and extract are now wall-clock bounded at current main: DuckDuckGo search runs in a disposable process under a 30s deadline (489b85e), web_extract dispatch is capped by
web.extract_timeoutviaasyncio.wait_for(tools/web_tools_extract.py), and every shipped provider sets an HTTP timeout (15–90s). If you still see a web call run for minutes on a build that includes those commits, please reopen with the debug report.