sendmessage SENDMESS-3 · child of SENDMESS-1 EPIC: is native SendMessage faster and cheaper (tokens/turns) than the hub? Backlog

Measure: turns per exchange, native vs hub (send -> receive -> reply -> receive)

Done high plpm-llmmsgsrv-cc

Count tool calls and model turns each side needs for one round trip. Hub path today: send (1 call) + poke/notification + read_unread (1 call) + send reply. Native path: SendMessage (1 call) + inbound arrives at next tool boundary (0 extra calls?) - verify whether a drain call is needed at all.

Sub-tickets

No sub-tickets.

+ Add sub-ticket

Activity

wi-cli-venus created · 7w ago

parent=#2850
wi-cli-venus completed · 7w ago

Measured - 1 turn + 1 tool call per side on both transports; no drain call needed on either. Turn count is not a differentiator.
wi-cli-venus commented · 7w ago

SM-R realistic-body re-measure (coder receiver rows): hub 1145 vs native 1064 tok/probe = 7.1% gap; small-id arm 7.3%. Double-render is capped (stub above ~130 chars). Sender-side gap 20%. Separate numbers per side.

Status

Priority

Assignee

Parent

Type

chore

Created

7w ago by wi-cli-venus

Updated

7w ago

Closed

2026-08-22 05:36