ton/pi_harness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(ai): support message-anchored tool loading (#6474)

Browse Source 
This adds cache-friendly dynamic tool loading anchored to tool results. Purely additive active-tool changes are recorded with `addedToolNames`, allowing supported Anthropic and OpenAI Responses models to load tool definitions at the point they become available instead of placing them in the cached prompt prefix.

It retains safe fallback behavior for unsupported models and non-additive changes but it will wipe caches.
This commit is contained in:
Armin Ronacher
2026-07-10 23:52:54 +02:00
committed by GitHub
parent d7a48d30a0
commit 3d8f74357c
19 changed files with 894 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/ai/CHANGELOG.md
+1
View File
@@ -10,6 +10,7 @@
### Added
- Added cache-friendly dynamic tool loading. `ToolResultMessage.addedToolNames` marks where tools from `Context.tools` became available; Anthropic and OpenAI Responses use native deferred loading so late tools stay out of the cached prefix, while other providers continue using `Context.tools` normally.
- Added a separate opt-in `max` thinking level, including native `xhigh` and `max` support for GPT-5.6 and Anthropic adaptive-thinking effort metadata matching Anthropic's documentation: `max` on all adaptive Claude models, native `xhigh` on Opus 4.7/4.8, Sonnet 5, and Fable 5 only.
- Added request-wide input-token pricing tiers to model cost metadata and usage cost calculation.