llm_gateway 0.8.0 → 0.8.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 5cd3b8b829adb3e5793ac604aa50c40bdd94546a5c2fc0b90125ca9a75efe4c2
-  data.tar.gz: 28e7211450893a23c2116d1cf96010fe5dd359c6203ecbde6802dda89f06f153
+  metadata.gz: c70b5fdd6e77e60cbe437a5698e078b7aa88fb41dca96311dcf9a58585238476
+  data.tar.gz: cb9dd1f8aa71c33751ddd270acb849d964ed137b6b3fdd6c96e072082bc6e488
 SHA512:
-  metadata.gz: f51eb7792fd0886403c9d3fec068dcd8f00d50c35df747e6735b470ccc170e8129bc8fe5ef18b8cd117a763e95b8cf9bc76c54364ba5c8b5f284f3e0b5ef69a5
-  data.tar.gz: c0de665f7e70b60814c8ea8185b34205de309736d4c578fee1d48de4990d76b4d437c1c1b74672d7f57c67b115e244e421f4b17882ea34cb875dcd5ed31dec59
+  metadata.gz: d21ab094dd5f600731e12541c6ba0a672eb8d0862c732ce2f5f2a415c394024ff87565dee5820e2d1dcec5ee665715d7beacc5eb5fb5a35d518e98b4c3c6c6e1
+  data.tar.gz: 97f4cabb1416beddc261fc849d23823722f9c36afb70843fcb1ef9e748e7f7eeec7f8bb4cfa0f92b25dbc864bf2c27e8751eeb9367f5c6233a906ac79d483219

data/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## [v0.8.1](https://github.com/Hyper-Unearthing/llm_gateway/tree/v0.8.1) (2026-06-27)
+
+[Full Changelog](https://github.com/Hyper-Unearthing/llm_gateway/compare/v0.8.0...v0.8.1)
+
+**Merged pull requests:**
+
+- fix: do not return empty assistant blocks [\#94](https://github.com/Hyper-Unearthing/llm_gateway/pull/94) ([billybonks](https://github.com/billybonks))
+- feat: when codex hits a rate limit pass all the information [\#93](https://github.com/Hyper-Unearthing/llm_gateway/pull/93) ([billybonks](https://github.com/billybonks))
+- Feat/fix proxy client [\#92](https://github.com/Hyper-Unearthing/llm_gateway/pull/92) ([billybonks](https://github.com/billybonks))
+
 ## [v0.8.0](https://github.com/Hyper-Unearthing/llm_gateway/tree/v0.8.0) (2026-06-06)
 
 [Full Changelog](https://github.com/Hyper-Unearthing/llm_gateway/compare/v0.7.0...v0.8.0)

data/README.md

@@ -916,6 +916,21 @@ puts result.content.select { |b| b.type == "text" }.map(&:text).join
 
 If your app refreshes tokens in the background, rebuild the adapter (or recreate client state) with the newest `access_token` before subsequent calls.
 
+### Codex rate-limit reset metadata
+
+OpenAI Codex usage-limit responses include reset information on `LlmGateway::Errors::RateLimitError`:
+
+```ruby
+begin
+  adapter.stream("Hello from OAuth auth", model: "gpt-5.4")
+rescue LlmGateway::Errors::RateLimitError => e
+  puts e.message                 # "The usage limit has been reached"
+  puts e.reset_after_seconds     # primary reset window, when available
+  puts e.reset_at                # Time for the primary reset, when available
+  puts e.rate_limit_info.inspect # full parsed Codex headers/body metadata
+end
+```
+
 ### Token refresh responsibility
 
 #### Library’s role (llm_gateway)

data/lib/llm_gateway/adapters/normalized_stream_accumulator.rb

@@ -330,7 +330,9 @@ module LlmGateway
       end
 
       def serialized_blocks
-        blocks.compact.map do |content_block|
+        blocks.compact.filter_map do |content_block|
+          next if blank_text_block?(content_block)
+
           if [ "tool_use", "server_tool_use" ].include?(content_block[:type])
             next content_block.merge(input: parse_tool_input(content_block[:input]).deep_symbolize_keys)
           end
@@ -348,6 +350,10 @@ module LlmGateway
         end
       end
 
+      def blank_text_block?(content_block)
+        content_block[:type] == "text" && content_block[:text].to_s.blank?
+      end
+
       def parse_tool_input(input)
         return {} if input.blank?
 

data/lib/llm_gateway/agents/harness.rb

@@ -128,6 +128,11 @@ module LlmGateway
       end
 
       def enqueue_or_run_message(message, queue, &block)
+        if session_manager.idle?
+          drain_queue(:steer)
+          drain_queue(:next_turn)
+          drain_queue(:follow_up)
+        end
         prepared_input = LlmGateway::Utils.deep_symbolize_keys(message)
         result = session_manager.start_or_enqueue_user_message(prepared_input, queue: queue) do
           compact_if_needed
@@ -135,6 +140,8 @@ module LlmGateway
         return if result == session_manager.class::MESSAGE_QUEUED
 
         begin
+
+
           continue(&block)
 
           loop do

data/lib/llm_gateway/agents/in_memory_session_manager.rb

@@ -45,6 +45,10 @@ module LlmGateway
         @state == :busy
       end
 
+      def idle?
+        @state == :idle
+      end
+
       def validate_queue!(queue)
         queue = queue.to_sym
         raise ArgumentError, "Invalid queue mode: #{queue}" unless QUEUES.include?(queue)

data/lib/llm_gateway/clients/openai.rb

@@ -1,5 +1,7 @@
 # frozen_string_literal: true
 
+require "time"
+
 require_relative "../base_client"
 
 module LlmGateway
@@ -179,6 +181,87 @@ module LlmGateway
         }
       end
 
+      def rate_limit_error_options(response, error)
+        info = rate_limit_info(response, error)
+        return {} if info.empty?
+
+        {
+          reset_at: info[:primary_reset_at] || info[:reset_at],
+          reset_after_seconds: info[:primary_reset_after_seconds] || info[:reset_after_seconds],
+          rate_limit_info: info
+        }
+      end
+
+      def rate_limit_info(response, error)
+        headers = response_headers(response)
+        codex_headers = headers.select { |key, _value| key.start_with?("x-codex-") }
+        info = {}
+
+        info[:provider] = "openai_codex" if codex_headers.any?
+        info[:error_type] = error["type"]
+        info[:plan_type] = headers["x-codex-plan-type"] || error["plan_type"]
+        info[:active_limit] = headers["x-codex-active-limit"]
+        info[:primary_used_percent] = integer_header(headers, "x-codex-primary-used-percent")
+        info[:secondary_used_percent] = integer_header(headers, "x-codex-secondary-used-percent")
+        info[:primary_window_minutes] = integer_header(headers, "x-codex-primary-window-minutes")
+        info[:secondary_window_minutes] = integer_header(headers, "x-codex-secondary-window-minutes")
+        info[:primary_over_secondary_limit_percent] = integer_header(headers, "x-codex-primary-over-secondary-limit-percent")
+        info[:primary_reset_after_seconds] = integer_header(headers, "x-codex-primary-reset-after-seconds")
+        info[:secondary_reset_after_seconds] = integer_header(headers, "x-codex-secondary-reset-after-seconds")
+        info[:primary_reset_at] = epoch_time(headers["x-codex-primary-reset-at"])
+        info[:secondary_reset_at] = epoch_time(headers["x-codex-secondary-reset-at"])
+        info[:credits_has_credits] = boolean_header(headers, "x-codex-credits-has-credits")
+        info[:credits_balance] = integer_header(headers, "x-codex-credits-balance")
+        info[:credits_unlimited] = boolean_header(headers, "x-codex-credits-unlimited")
+        info[:reset_after_seconds] = integer_value(error["resets_in_seconds"]) || retry_after_seconds(headers["retry-after"])
+        info[:reset_at] = epoch_time(error["resets_at"]) || retry_after_time(headers["retry-after"])
+
+        info.compact
+      end
+
+      def response_headers(response)
+        headers = {}
+        response.each_header { |key, value| headers[key.downcase] = value }
+        headers
+      end
+
+      def integer_header(headers, key)
+        integer_value(headers[key])
+      end
+
+      def integer_value(value)
+        return nil if value.nil?
+
+        Integer(value)
+      rescue ArgumentError, TypeError
+        nil
+      end
+
+      def boolean_header(headers, key)
+        case headers[key].to_s.downcase
+        when "true" then true
+        when "false" then false
+        end
+      end
+
+      def epoch_time(value)
+        integer = integer_value(value)
+        Time.at(integer) if integer
+      end
+
+      def retry_after_seconds(value)
+        integer_value(value)
+      end
+
+      def retry_after_time(value)
+        seconds = retry_after_seconds(value)
+        return Time.now + seconds if seconds
+
+        Time.httpdate(value)
+      rescue ArgumentError, TypeError
+        nil
+      end
+
       def handle_client_specific_errors(response, error)
         # OpenAI uses 'code' instead of 'type' for error codes
         error_code = error["code"]
@@ -190,7 +273,7 @@ module LlmGateway
 
         case response.code.to_i
         when 429
-          raise Errors::RateLimitError.new(error_message, error_code)
+          raise Errors::RateLimitError.new(error_message, error_code, **rate_limit_error_options(response, error))
         when 503
           raise Errors::OverloadError.new(error_message, error_code)
         end

data/lib/llm_gateway/errors.rb

@@ -19,7 +19,16 @@ module LlmGateway
     class NotFoundError < ClientError; end
     class ConflictError < ClientError; end
     class UnprocessableEntityError < ClientError; end
-    class RateLimitError < ClientError; end
+    class RateLimitError < ClientError
+      attr_reader :reset_at, :reset_after_seconds, :rate_limit_info
+
+      def initialize(message = nil, code = nil, reset_at: nil, reset_after_seconds: nil, rate_limit_info: {})
+        @reset_at = reset_at
+        @reset_after_seconds = reset_after_seconds
+        @rate_limit_info = rate_limit_info
+        super(message, code)
+      end
+    end
     class InternalServerError < ClientError; end
     class APIStatusError < ClientError; end
     class APITimeoutError < ClientError; end

data/lib/llm_gateway/proxy/adapter.rb

@@ -3,10 +3,11 @@
 module LlmGateway
   module Proxy
     class Adapter
-      attr_reader :client
+      attr_reader :client, :provider_key
 
-      def initialize(client)
+      def initialize(client, provider_key: nil)
         @client = client
+        @provider_key = provider_key
       end
 
       def stream(message, tools: nil, system: nil, **options, &block)

data/lib/llm_gateway/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LlmGateway
-  VERSION = "0.8.0"
+  VERSION = "0.8.1"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: llm_gateway
 version: !ruby/object:Gem::Version
-  version: 0.8.0
+  version: 0.8.1
 platform: ruby
 authors:
 - billybonks
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2026-06-06 00:00:00.000000000 Z
+date: 2026-06-27 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: dry-struct