From f4aa2d01e581bc06e3b0ac360a46058db5ba7596 Mon Sep 17 00:00:00 2001 From: "Jakub A. W" Date: Tue, 28 Jul 2026 11:24:16 +0200 Subject: [PATCH] docs(pro): Responses compression coverage and revision-body toggle Token compression now covers POST /v1/responses (GoModel-pro#7): document the endpoint, the structural exemptions there (replayed reasoning items and annotated text parts), and the new LOGGING_LOG_REVISION_BODIES flag (GoModel#603) that trades the stored rewritten-body copy for roughly half the audit growth. Co-Authored-By: Claude Fable 5 --- docs/pro.mdx | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-) diff --git a/docs/pro.mdx b/docs/pro.mdx index 3453115d..a6a9fad0 100644 --- a/docs/pro.mdx +++ b/docs/pro.mdx @@ -47,6 +47,10 @@ Measured on a corpus of real agent conversations at default settings: - tool-call arguments, non-text content parts, and unknown JSON fields - first occurrences — only later duplicates are replaced +- on `/v1/responses`: replayed reasoning items (the API requires them back + verbatim when `store` is off, so `PRO_COMPRESSION_DROP_REASONING` has no + effect there) and text parts carrying citation annotations, whose + start/end offsets a rewrite would break - anything, if estimated savings fall below the threshold; the request passes through byte-identical @@ -55,7 +59,9 @@ always compresses the same way and the provider's prompt-cache prefix stays stable as a conversation grows. It fails open: if the engine errors, the request is forwarded unchanged. -Covered endpoints: `POST /v1/chat/completions` and `POST /v1/messages`. +Covered endpoints: `POST /v1/chat/completions`, `POST /v1/messages`, and +`POST /v1/responses` — including function and custom tool-call outputs on +the Responses API. ## Turn it off @@ -122,7 +128,10 @@ X-GoModel-Pro-Compression-Blocks: 12 Audit entries keep the **original** client request plus a revision chain showing exactly what each rewriter changed, so the full before/after is -always recoverable. +always recoverable. Storing the rewritten body copy is governed by +`LOGGING_LOG_REVISION_BODIES` (default `true`): turning it off keeps the +revision metadata — rewriter, sizes, tokens saved, the block report — while +roughly halving audit storage per compressed request. **Aggregate** — Prometheus counters on `/metrics`, all prefixed `gomodel_pro_compression_`, and the dashboard **Usage** page shows a "Pro