hermes-agent/agent/transports
Soju06 7b3dcee928 feat(cache): persist the exact bytes sent to the API in an api_content sidecar
The first LLM call of every gateway turn gets ~0% provider prompt-cache
hit rate (in-turn calls: 97-99%) because the bytes sent for a turn's
user message are not the bytes replayed next turn: memory-prefetch and
pre_llm_call context are injected into the API copy only, the #48677
persist override writes cleaned content to the DB row, and
get_messages_as_conversation sanitize/strips user and assistant content
on load. Any of these diverges the request prefix at that message and
re-prefills everything after it — measured 27.9s for the first call vs
2.4-5.8s cached at a median ~156k-token context.

Persist what you send: a nullable messages.api_content column stores the
exact content string sent to the API when it differs from the clean
stored content, and replay substitutes it verbatim (no sanitize, no
strip). The injection composition lives in one helper
(turn_context.compose_user_api_content); the turn prologue stamps its
output onto the live user message, the api_messages build sends the
stamped bytes, and every outgoing copy pops the field so it never
reaches a provider. The crash-resilience user-turn persist moves after
prefetch/pre_llm_call so the user row is written once with its final
sidecar; _ensure_db_session stays before preflight compression (session
rotation needs the parent row under PRAGMA foreign_keys=ON). The
current-turn index trackers are re-anchored after compaction rebuilds
the message list, in-place preflight compaction backfills the stamp onto
the already-inserted row, and gateway replay forwards the sidecar only
when the replay pipeline did not rewrite the content. Rewrite paths that
would leave stale bytes (historical image strip, merge-summary-into-
tail, consecutive-user repair merge, stale-confirmation redaction) drop
the sidecar; the chat-completions transport and the max-iterations
summary path strip it defensively. codex_app_server and MoA turns are
excluded from stamping because their wire bytes differ from the
composition. A missing or dropped sidecar degrades to today's behavior
(one cache-boundary miss), never to wrong content.
2026-07-19 08:25:35 +05:30
..
__init__.py feat: provider modules — ProviderProfile ABC, 33 providers, fetch_models, transport single-path 2026-05-05 13:40:01 -07:00
anthropic.py fix(anthropic): also normalize MCP-server tool names to mcp__ on OAuth wire 2026-06-17 13:20:29 +05:30
base.py feat: add transport ABC + AnthropicTransport wired to all paths 2026-04-21 01:27:01 -07:00
bedrock.py feat: add BedrockTransport + wire all Bedrock transport paths 2026-04-21 20:58:37 -07:00
chat_completions.py feat(cache): persist the exact bytes sent to the API in an api_content sidecar 2026-07-19 08:25:35 +05:30
codex.py fix(codex): harden final cache-key boundaries 2026-07-17 13:48:41 -07:00
codex_app_server.py fix(security): strip dynamic Hermes secrets from all subprocess spawn env 2026-07-01 14:37:22 +05:30
codex_app_server_session.py fix(codex): forward drained notifications to on_event during approval roundtrips 2026-07-17 06:49:47 -07:00
codex_event_projector.py feat(mcp): adopt mcp__server__tool naming convention 2026-07-05 13:40:21 -07:00
hermes_tools_mcp_server.py chore(mcp): drop stale input_schema comment on add_tool call 2026-07-14 21:36:32 -07:00
types.py fix(anthropic): preserve interleaved thinking/tool_use block order on replay 2026-06-10 20:45:16 -07:00