MoA aggregator prompt cache billing and routing fixes
NousResearch/hermes-agent · cluster 112358
Data as of
Nothing open on GitHub. Fix merged in #113175. Check it covers your case.
The short answer
- Known?
- Yes. 6 related items grouped here, 1 issue already closed.
- Fix event
- A fix was merged to main in PR #113175. Check it covers your case. No release data is tracked.
- Tracker
- The maintainers' tracker marks this
resolved, deciding PR #115606. That is separate from the fix event. - Still open
- Nothing open on GitHub.
Multiple fixes targeting how the MoA aggregator handles prompt cache identity, routing, and persistence across tool steps and iterations. A related issue documents unexpected re-billing behavior when tool history grows and cache fails to stay pinned to the system prompt.
Open issues0
Open PRs0
Fixes on record
-
#113175 fix(moa): aggregator prompt cache no longer collapses to the system prompt on every tool step (#112358)
files:
agent/moa_loop.py,tests/agent/test_failover_identity.py,tests/agent/test_moa_aggregator_prefix_stability.py,website/docs/user-guide/features/mixture-of-agents.md -
#113784 fix(moa): Anthropic-wire aggregator keeps the cache prefix through the task on iteration 1 (closes remaining atoms of #112358)
files:
agent/anthropic_message_convert.py,agent/moa_loop.py,tests/agent/test_anthropic_adapter.py,tests/agent/test_moa_aggregator_prefix_stability.py,website/docs/user-guide/features/mixture-of-agents.md -
#115606 fix(moa): aggregator recovers from strict-alternation 400s per destination (closes remaining atoms of #112358)
files:
agent/chat_completion_helpers.py,agent/error_classifier.py,agent/moa_alternation.py,agent/moa_loop.py,agent/turn_failure_copy.py,tests/agent/test_error_classifier.py,tests/agent/test_moa_alternation_recovery.py,website/docs/user-guide/features/mixture-of-agents.md
2 PRs closed without merging
From the close comments: done elsewhere 2
Closed issues · 1
- Closed#112358MoA aggregator: prompt cache hits stay pinned to the system prompt; growing tool history re-billed every call (~$147 in 49 min)
How this cluster was grouped
-
Link evidence (
structural links throughout): every member is connected by structural evidence: shared changed files, or a closing/superseded reference. -
Pairwise link density
0.33: the share of item pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36.1.00means every pair is linked.
Pick up a Lab task
Check whether a fix covers the open issues here, with your own agent. Bring back evidence for maintainers.
Open the Lab