ask-llm-providers 0.13.2 → 0.13.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 3ce937df3ae9db671946c8d5c3d837498fd521066bb45996e595dac977ff9f24
-  data.tar.gz: 9702d0c8aea1b61aa24df7e8bb76c5cc19e576ce926076fb274bdb095691b666
+  metadata.gz: a1561002d3a643aa61fa750f7074a1a362d45a108f9ea3c44b8d21f0704e6907
+  data.tar.gz: d4702a9c31b9a3dda3bea771bd5722a31ccfa5443ebcd4b64a235cafad319d96
 SHA512:
-  metadata.gz: 3ea28d9f251de830592fb92f7d78c71d9b7983dbfc69aa29617ef06df5d598edfa5d9e5a9c8f386a2f9b2d9be22cb76a9c54594c7710acfea17779aa7de9c16d
-  data.tar.gz: d67134217af9ddb29c91b8dd4c64cb1bbe68f2fd79da31bf0bac7a7b64abb681ca0f34e11e7327d587de13c531b722466075cb5e10f35c302acf174af4d3e17b
+  metadata.gz: 2e8c1f0e2f652dbd7031a27278f6e6ead1b23de697450ed14aa0882d2225c0d7392c9de0657504507740199112289ba70adc0ebb33b23427ffb4bc0b3106ef94
+  data.tar.gz: 0154ff38051805e13c36fa9c7408ac7fbf5bb38d16362ebe215a9e3e39385b24db822afd2c872cb2a41424993ce4546fc00bf547af927936a2023d6381ab6ef6

data/lib/ask/llm/models.json

@@ -5076,6 +5076,28 @@
       "output": 0
     }
   },
+  {
+    "id": "deepseek/deepseek-v4.1-flash",
+    "name": "DeepSeek V4.1 Flash",
+    "provider": "commandcode",
+    "context_window": 1000000,
+    "capabilities": [
+      "function_calling",
+      "streaming"
+    ],
+    "modalities": {
+      "input": [
+        "text"
+      ],
+      "output": [
+        "text"
+      ]
+    },
+    "pricing": {
+      "input": 0,
+      "output": 0
+    }
+  },
   {
     "id": "deepseek/deepseek-v4-flash",
     "name": "DeepSeek V4 Flash (latest)",

data/lib/ask/llm/openai_compatible.rb

@@ -76,6 +76,11 @@ module Ask
                        capabilities: { chat: true, streaming: true, tool_calls: true } },
 
       opencode_go:   { api_base: "https://opencode.ai/zen/go/v1",                 api_key_env: "OPENCODE_GO_API_KEY",
+                        # The Go gateway routes by session and refuses a request
+                        # without this header (400 MissingSessionID) — the whole
+                        # call fails, not just its efficiency. A per-client
+                        # UUID keeps one agent's requests on one route.
+                        session_header: "x-opencode-session",
                         capabilities: { chat: true, streaming: true, tool_calls: true } },
 
       github_copilot: { api_base: "https://api.githubcopilot.com",                  api_key_env: "GITHUB_COPILOT_TOKEN",

data/lib/ask/llm/version.rb

@@ -2,6 +2,6 @@
 
 module Ask
   module LLM
-    VERSION = "0.13.2"
+    VERSION = "0.13.4"
   end
 end

data/lib/ask/provider/openai.rb

@@ -496,16 +496,20 @@ module Ask
       def chat_stream(payload, model, &block)
         stream = Ask::Stream.new
         init_sse_buffer
+        # The response body is consumed by the stream callback below, so on
+        # failure `resp.body` is empty — for exactly the responses whose text
+        # matters most. Keeping the first chunk of what actually arrived is
+        # what lets an error carry the provider's own words.
+        raw = +""
         @http.post("chat/completions") do |req|
           req.body = payload.merge(stream: true)
-          req.options.on_data = proc { |data, _bytes, _env| parse_stream(data, stream, model, &block) }
+          req.options.on_data = proc { |data, _bytes, _env|
+            raw << data if raw.bytesize < MAX_ERROR_BODY
+            parse_stream(data, stream, model, &block)
+          }
         end.tap { |resp|
           unless resp.success?
-            err_body = case resp.body
-                       when Hash then resp.body
-                       when String then (JSON.parse(resp.body) rescue { "error" => { "message" => "HTTP #{resp.status}: #{resp.body[0..200]}" } })
-                       else { "error" => { "message" => "HTTP #{resp.status}: empty response body" } }
-                       end
+            err_body = stream_error_body(resp, raw)
             err_body["error"] ||= {}
             err_body["error"]["_status"] = resp.status
             raise LLM::HTTP.map_error(resp.status, err_body, provider: "OpenAI")
@@ -515,6 +519,34 @@ module Ask
         stream
       end
 
+      # How much of a failed response to keep for the error message. Enough
+      # for a JSON error document, not enough to hold a whole streamed body
+      # in memory.
+      MAX_ERROR_BODY = 4096
+
+      # The provider's error, from wherever it survived.
+      #
+      # Faraday hands the body to `on_data` as it arrives, so a response that
+      # failed has nothing left in `resp.body` — which is how a specific,
+      # actionable 400 ("Request is missing x-opencode-session") reached the
+      # operator as "empty response body" and cost an afternoon. Fall back to
+      # the buffered text, then to whatever we do have.
+      def stream_error_body(resp, raw)
+        body = resp.body
+        return body if body.is_a?(Hash)
+
+        text = body.to_s
+        text = raw if text.strip.empty?
+        parsed = begin
+          JSON.parse(text)
+        rescue
+          nil
+        end
+        return parsed if parsed.is_a?(Hash) && parsed["error"]
+
+        {"error" => {"message" => "HTTP #{resp.status}: #{text.to_s.strip[0..400]}"}}
+      end
+
       def parse_tool_calls(calls)
         return nil unless calls&.any?
 

data/lib/ask/provider/openai_compatible.rb

@@ -1,5 +1,7 @@
 # frozen_string_literal: true
 
+require "securerandom"
+
 module Ask
   module Providers
     # Single class for all OpenAI-compatible providers.
@@ -57,9 +59,23 @@ module Ask
         if (extra = @compat_cfg[:extra_headers])
           extra.each { |k, v| h[k] = v }
         end
+        # A gateway that routes by session needs an id on every request, and
+        # a stable one per client so requests land on the same route. This is
+        # why the failure mode is a hard 400 rather than a slow call: the
+        # gateway cannot place the request at all.
+        if (header = @compat_cfg[:session_header])
+          h[header] = session_id
+        end
         h
       end
 
+      # One id per provider instance — a conversation's requests stay on one
+      # route, which is what the header is for. Memoized so every request
+      # from this client carries the same value.
+      def session_id
+        @session_id ||= SecureRandom.uuid
+      end
+
       def format_messages(messages)
         result = super
         if @compat_cfg[:reasoning_content]

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ask-llm-providers
 version: !ruby/object:Gem::Version
-  version: 0.13.2
+  version: 0.13.4
 platform: ruby
 authors:
 - Kaka Ruto