You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Allow read-only tools during fan-out and preserve tool history in conversation
Two fixes for "providers not receiving tool execution results":
1. Fan-out now supports read-only tools (file_read). Providers can inspect
the codebase during their initial response instead of giving generic
advice. Write/exec tools remain synthesis-only. Fan-out tool calls are
executed inline with up to 3 re-query rounds per provider.
2. Tool calls and results are now preserved as structured messages in the
conversation state (assistant+ToolCalls → RoleTool results → follow-up).
Previously flattened to a text blob, losing tool context for subsequent
turns. ToolLoop.RunWithMessages returns the appended messages for the
caller to store in conv.
Copy file name to clipboardExpand all lines: CHANGELOG.md
+8Lines changed: 8 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -6,6 +6,14 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/), and this
6
6
7
7
## [Unreleased]
8
8
9
+
## [1.13.0] - 2026-03-23
10
+
11
+
### Added
12
+
-**Read-only tools during fan-out**: Providers can now use `file_read` during fan-out to inspect the codebase before responding, instead of giving generic advice. Write and exec tools remain synthesis-only.
13
+
14
+
### Fixed
15
+
-**Tool results lost from conversation history**: Tool calls and results are now preserved as structured messages in the conversation state (assistant+tool_calls → tool results → follow-up). Previously they were flattened to a single text blob, so providers lost tool context on subsequent turns.
Copy file name to clipboardExpand all lines: README.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -29,7 +29,7 @@ Every AI model has blind spots. Claude might excel at architecture, GPT at debug
29
29
30
30
### Multi-Provider Consensus
31
31
32
-
Query every configured LLM simultaneously. Responses fan out in parallel (latency = slowest provider, not sum of all), and your designated **primary** model synthesizes them into a single answer — identifying areas of agreement, unique insights, and errors.
32
+
Query every configured LLM simultaneously. Responses fan out in parallel (latency = slowest provider, not sum of all), and your designated **primary** model synthesizes them into a single answer — identifying areas of agreement, unique insights, and errors. Providers can read files during fan-out to give codebase-aware answers.
0 commit comments