legion-llm 0.14.8 → 0.14.9

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: b12fc37520233aba6bc8b063e0334e5e1abb74970c5ba6a2efd9094bbd3414cf
-  data.tar.gz: f52cdb8f4cc7c7c2e67e0a6c8caa55fb842f736418955c4ec20d89a0b677212a
+  metadata.gz: f67a1a83e7213a2e50437b07e71afd74302c2b1da8229708ec8d99506d45354e
+  data.tar.gz: fa1dcc7ae0d4b64dd2716271dd5c50778ff178fa3987704b16302c00b75029ab
 SHA512:
-  metadata.gz: 598310ad96bad4d54ac02802a6d981294d5fb1faec31390fa6375e378f936f47f2336027255e0ca5c49f6db199b606528c8e96b70112325e41ccbb346245db11
-  data.tar.gz: 68c89f512cea238a856fdc5930155a7018cb095573f0de5de50bce8f03bb662c2b67abdd0f653a7f201139e3412806beef631f1f46dd80f065c18c0a496c489b
+  metadata.gz: b3c1802c7bc32daa15d15c0fc8b16923d8fac48531adc5bcfb2997eec6234534166b85c43e9262774dedb63e28a7a62c148614bdac53e00c15fe44598770823b
+  data.tar.gz: cc6ae96b0bfc1e1e54298b028334d1bd1f19e18147ac0d7a7d61bb849c86e157ff26b986082be3027ba4199570481d91b960ce6eb1d82b2ec5aa66d787d3a9b7

data/CHANGELOG.md

@@ -1,5 +1,18 @@
 # Legion LLM Changelog
 
+## [0.14.9] - 2026-06-25
+
+### Fixed
+
+- Routing token estimate (`estimate_request_tokens`) now includes system prompt, tool definitions, and thinking config — matching what `enforce_final_context_budget!` checks at dispatch time. Previously only counted messages, causing large-context requests (e.g. Copilot Chat with 81 tools + system prompt) to route to small-context models and fail with ContextOverflow at dispatch.
+- `resolve_routing_state` always consults `request_lane` when Inventory has lanes, removing the gate that required explicit routing flags. Enables weight-based lane selection for all callers including `Legion::LLM.chat`.
+
+### Added
+
+- `:subcall` profile in `Inference::Profile` — skips all pipeline steps except routing, provider_call, and metering. Used for in-pipeline LLM sub-calls (debate, lex-knowledge, lex-apollo) to avoid recursive full-pipeline execution.
+- `Executor.new(request, skip_pipeline: true)` kwarg forces `:subcall` profile without requiring caller identity manipulation.
+- `dispatch_chat` unified path — builds a `Request` via `from_chat_args` and runs the Executor directly, detecting in-pipeline context via `Thread.current[:legion_llm_in_pipeline]`.
+
 ## [0.14.8] - 2026-06-25
 
 ### Fixed

data/lib/legion/llm/inference/executor/routing.rb

@@ -199,14 +199,15 @@ module Legion
           end
 
           def estimate_request_tokens
-            # Estimate total tokens from current request messages + conversation history.
-            # This is used by the router to exclude models whose context window can't fit.
             all_messages = []
             all_messages.concat(@enrichments['context:conversation_history'] || [])
             all_messages.concat(@request.messages || [])
-            return 0 if all_messages.empty?
 
-            estimate_message_tokens(all_messages)
+            estimated = all_messages.empty? ? 0 : estimate_message_tokens(all_messages)
+            estimated += ((@request.system || '').length / 4.0).ceil
+            estimated += estimate_tool_token_budget if @request.tools&.any?
+            estimated += (Legion::JSON.dump(@request.thinking).length / 3.5).ceil if @request.thinking.is_a?(Hash) && @request.thinking.any?
+            estimated
           end
 
           def chain_required_capabilities
@@ -367,11 +368,7 @@ module Legion
 
           def resolve_routing_state(state)
             return state unless defined?(Router)
-
-            explicit_route = state[:provider_explicit] || state[:instance_explicit] || state[:tier_explicit]
-            auto_route = state[:auto_route] == true
-            intent_route = state[:intent_explicit] && state[:intent] && Router.routing_enabled?
-            return state unless explicit_route || auto_route || intent_route
+            return state unless Legion::LLM::Inventory.lanes.any?
 
             resolution = routing_resolution_for(state)
             return state unless resolution

data/lib/legion/llm/version.rb

@@ -2,6 +2,6 @@
 
 module Legion
   module LLM
-    VERSION = '0.14.8'
+    VERSION = '0.14.9'
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: legion-llm
 version: !ruby/object:Gem::Version
-  version: 0.14.8
+  version: 0.14.9
 platform: ruby
 authors:
 - Esity