llm.rb 5.1.0 → 5.2.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 56ddedb75f6c791cc42bca736bc62360ba4850a3a204f9a82288e8c6ea977eeb
-  data.tar.gz: 3881b731dacd921e258eac954c4468d052e673e48ad53c63ae1a246973c84d33
+  metadata.gz: 8f9bdef0c733225e44dcf39d75e3397974122bfeb5e705a0797067242fd5c966
+  data.tar.gz: 567cc793e1e095e481abf5ef797a6fcb26a04faeed91855c234e531b78e3544a
 SHA512:
-  metadata.gz: a8838f57a1232afc42448d28a0f3f7b8907c2a527be284579b3af56e398edda50d7cc02a8dda2794c65096699831058494cccae3f8a116b538f76bc42127eba8
-  data.tar.gz: ef49e8046b4aab4e59b252ffdbf16135673d227b4124bf45bf5c31856c49822168182928b56fdc7428d2056dbae7c211fac0d6d8ef3eba48a7aca47420bb96e7
+  metadata.gz: 3d7e026b308228787d2f6ead8de197f847b644f7ba5bd0a1d679270a66e5c0e48c74a7d9494a86613bd061a42c2fd56ff270f73f8f3a627bc213dd7d57de788d
+  data.tar.gz: 8586a02d0345e7259f80b32e688a0ad531e28e3d15c8519781647ee1f77f23b6ecdaa9b28ed40efdea138c138511c7c4f88986ed7d646fb562e9faf7e2db687f

data/CHANGELOG.md

@@ -2,8 +2,68 @@
 
 ## Unreleased
 
+Changes since `v5.2.1`.
+
+## v5.2.1
+
+Changes since `v5.2.0`.
+
+This release tightens the streamed queue fix from `v5.2.0` for concurrent
+workloads. Request-local streams now stay bound long enough for `wait` to
+drain queued work and then clear cleanly so later waits fall back to the
+context's configured stream.
+
+### Fix
+
+* **Reset request-local streams after `wait` drains queued work** <br>
+  Keep per-call `stream:` bindings alive through `LLM::Context#wait` so
+  queued streamed tool work still resolves correctly, then clear the
+  request-local stream after the wait completes to avoid leaking it into
+  later turns.
+
+## v5.2.0
+
 Changes since `v5.1.0`.
 
+This release adds current DeepSeek V4 support through refreshed provider
+metadata, including `deepseek-v4-flash` and `deepseek-v4-pro`, while fixing
+request-local queue handling for concurrent streamed workloads so `wait` and
+interruption use the active per-call stream correctly.
+
+### Change
+
+* **Add `LLM::MCP#run` for scoped MCP client lifecycle** <br>
+  Add `LLM::MCP#run` so MCP clients can be started for the duration of a
+  block and then stopped automatically, which simplifies the usual
+  `start`/`stop` pattern in examples and application code.
+
+* **Refresh provider model metadata** <br>
+  Add current DeepSeek and OpenAI model metadata to `data/` and update the
+  Google Gemma model entry to match the current provider naming.
+
+### Fix
+
+* **Reject unsupported DeepSeek multimodal prompt objects early** <br>
+  Raise `LLM::PromptError` for `image_url`, `local_file`, and
+  `remote_file` in DeepSeek chat requests instead of sending invalid
+  OpenAI-compatible payloads that the provider rejects at runtime.
+
+* **Preserve DeepSeek reasoning content across tool turns** <br>
+  Replay `reasoning_content` when serializing prior assistant messages for
+  DeepSeek chat completions, so thinking-mode tool calls can continue into
+  follow-up requests without triggering invalid request errors.
+
+* **Default DeepSeek to `deepseek-v4-flash`** <br>
+  Change `LLM::DeepSeek#default_model` to `deepseek-v4-flash` so new
+  contexts and default provider usage align with the current preferred chat
+  model.
+
+* **Use per-call streams when waiting on streamed tool work** <br>
+  Track request-local streams bound through `talk(..., stream:)` and
+  `respond(..., stream:)` so `LLM::Context#wait` and interruption-aware
+  queue handling use the active stream instead of falling back to pending
+  function spawning.
+
 ## v5.1.0
 
 Changes since `v5.0.0`.

data/README.md

@@ -4,7 +4,7 @@
 <p align="center">
   <a href="https://0x1eef.github.io/x/llm.rb?rebuild=1"><img src="https://img.shields.io/badge/docs-0x1eef.github.io-blue.svg" alt="RubyDoc"></a>
   <a href="https://opensource.org/license/0bsd"><img src="https://img.shields.io/badge/License-0BSD-orange.svg?" alt="License"></a>
-  <a href="https://github.com/llmrb/llm.rb/tags"><img src="https://img.shields.io/badge/version-5.1.0-green.svg?" alt="Version"></a>
+  <a href="https://github.com/llmrb/llm.rb/tags"><img src="https://img.shields.io/badge/version-5.2.1-green.svg?" alt="Version"></a>
 </p>
 
 ## About
@@ -261,13 +261,17 @@ Remote MCP tools and prompts are not bolted on as a separate integration
 stack. They adapt into the same tool and prompt path used by local tools,
 skills, contexts, and agents.
 
+Use `mcp.run do ... end` for scoped work where the client should start and
+stop around one block. Use `mcp.start` and `mcp.stop` directly when you need
+finer sequential control across several steps before shutting the client down.
+
 ```ruby
-begin
-  mcp = LLM::MCP.http(url: "https://api.githubcopilot.com/mcp/").persistent
-  mcp.start
+mcp = LLM::MCP.http(
+  url: "https://api.githubcopilot.com/mcp/",
+  headers: {"Authorization" => "Bearer #{ENV.fetch("GITHUB_PAT")}"}
+).persistent
+mcp.run do
   ctx = LLM::Context.new(llm, tools: mcp.tools)
-ensure
-  mcp.stop
 end
 ```
 
@@ -281,12 +285,17 @@ Go's context package. In fact, llm.rb is heavily inspired by Go but with a Ruby
 twist.
 
 ```ruby
+require "llm"
+require "io/console"
+
+llm = LLM.openai(key: ENV["KEY"])
 ctx = LLM::Context.new(llm, stream: $stdout)
 worker = Thread.new do
   ctx.talk("Write a very long essay about network protocols.")
 rescue LLM::Interrupt
   puts "Request was interrupted!"
 end
+
 STDIN.getch
 ctx.interrupt!
 worker.join
@@ -615,9 +624,10 @@ require "io/console"
 
 llm = LLM.openai(key: ENV["KEY"])
 ctx = LLM::Context.new(llm, stream: $stdout)
-
 worker = Thread.new do
   ctx.talk("Write a very long essay about network protocols.")
+rescue LLM::Interrupt
+  puts "Request was interrupted!"
 end
 
 STDIN.getch
@@ -695,7 +705,7 @@ puts ticket.talk("How do I rotate my API key?").content
 
 #### MCP
 
-This example uses [`LLM::MCP`](https://0x1eef.github.io/x/llm.rb/LLM/MCP.html) over HTTP so remote GitHub MCP tools run through the same `LLM::Context` tool path as local tools. See the [deepdive (web)](https://0x1eef.github.io/x/llm.rb/file.deepdive.html) or [deepdive (markdown)](resources/deepdive.md) for more examples.
+This example uses [`LLM::MCP`](https://0x1eef.github.io/x/llm.rb/LLM/MCP.html) over HTTP so remote GitHub MCP tools run through the same `LLM::Context` tool path as local tools. It expects a GitHub token in `ENV["GITHUB_PAT"]`. See the [deepdive (web)](https://0x1eef.github.io/x/llm.rb/file.deepdive.html) or [deepdive (markdown)](resources/deepdive.md) for more examples.
 
 ```ruby
 require "llm"
@@ -707,13 +717,24 @@ mcp = LLM::MCP.http(
   headers: {"Authorization" => "Bearer #{ENV.fetch("GITHUB_PAT")}"}
 ).persistent
 
-begin
-  mcp.start
+mcp.start
+ctx = LLM::Context.new(llm, stream: $stdout, tools: mcp.tools)
+ctx.talk("Pull information about my GitHub account.")
+ctx.talk(ctx.call(:functions)) while ctx.functions.any?
+mcp.stop
+```
+
+For scoped work, `mcp.run do ... end` is shorter and handles cleanup for you:
+
+```ruby
+mcp = LLM::MCP.http(
+  url: "https://api.githubcopilot.com/mcp/",
+  headers: {"Authorization" => "Bearer #{ENV.fetch("GITHUB_PAT")}"}
+).persistent
+mcp.run do
   ctx = LLM::Context.new(llm, stream: $stdout, tools: mcp.tools)
   ctx.talk("Pull information about my GitHub account.")
   ctx.talk(ctx.call(:functions)) while ctx.functions.any?
-ensure
-  mcp.stop
 end
 ```
 

data/data/deepseek.json

@@ -70,6 +70,74 @@
         "context": 128000,
         "output": 64000
       }
+    },
+    "deepseek-v4-flash": {
+      "id": "deepseek-v4-flash",
+      "name": "DeepSeek V4 Flash",
+      "family": "deepseek-flash",
+      "attachment": false,
+      "reasoning": true,
+      "tool_call": true,
+      "interleaved": {
+        "field": "reasoning_content"
+      },
+      "structured_output": true,
+      "temperature": true,
+      "knowledge": "2025-05",
+      "release_date": "2026-04-24",
+      "last_updated": "2026-04-24",
+      "modalities": {
+        "input": [
+          "text"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": true,
+      "cost": {
+        "input": 0.14,
+        "output": 0.28,
+        "cache_read": 0.028
+      },
+      "limit": {
+        "context": 1000000,
+        "output": 384000
+      }
+    },
+    "deepseek-v4-pro": {
+      "id": "deepseek-v4-pro",
+      "name": "DeepSeek V4 Pro",
+      "family": "deepseek-thinking",
+      "attachment": false,
+      "reasoning": true,
+      "tool_call": true,
+      "interleaved": {
+        "field": "reasoning_content"
+      },
+      "structured_output": true,
+      "temperature": true,
+      "knowledge": "2025-05",
+      "release_date": "2026-04-24",
+      "last_updated": "2026-04-24",
+      "modalities": {
+        "input": [
+          "text"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": true,
+      "cost": {
+        "input": 1.74,
+        "output": 3.48,
+        "cache_read": 0.145
+      },
+      "limit": {
+        "context": 1000000,
+        "output": 384000
+      }
     }
   }
 }

data/data/google.json

@@ -1058,6 +1058,32 @@
         "output": 8192
       }
     },
+    "gemma-4-26b-a4b-it": {
+      "id": "gemma-4-26b-a4b-it",
+      "name": "Gemma 4 26B",
+      "family": "gemma",
+      "attachment": false,
+      "reasoning": true,
+      "tool_call": true,
+      "structured_output": true,
+      "temperature": true,
+      "release_date": "2026-04-02",
+      "last_updated": "2026-04-02",
+      "modalities": {
+        "input": [
+          "text",
+          "image"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": true,
+      "limit": {
+        "context": 256000,
+        "output": 8192
+      }
+    },
     "gemini-2.5-flash-lite": {
       "id": "gemini-2.5-flash-lite",
       "name": "Gemini 2.5 Flash Lite",
@@ -1093,32 +1119,6 @@
         "output": 65536
       }
     },
-    "gemma-4-26b-it": {
-      "id": "gemma-4-26b-it",
-      "name": "Gemma 4 26B",
-      "family": "gemma",
-      "attachment": false,
-      "reasoning": true,
-      "tool_call": true,
-      "structured_output": true,
-      "temperature": true,
-      "release_date": "2026-04-02",
-      "last_updated": "2026-04-02",
-      "modalities": {
-        "input": [
-          "text",
-          "image"
-        ],
-        "output": [
-          "text"
-        ]
-      },
-      "open_weights": true,
-      "limit": {
-        "context": 256000,
-        "output": 8192
-      }
-    },
     "gemini-2.5-flash-image-preview": {
       "id": "gemini-2.5-flash-image-preview",
       "name": "Gemini 2.5 Flash Image (Preview)",

data/data/openai.json

@@ -195,6 +195,61 @@
         "output": 16384
       }
     },
+    "gpt-5.5": {
+      "id": "gpt-5.5",
+      "name": "GPT-5.5",
+      "family": "gpt",
+      "attachment": true,
+      "reasoning": true,
+      "tool_call": true,
+      "structured_output": true,
+      "temperature": false,
+      "knowledge": "2025-12-01",
+      "release_date": "2026-04-23",
+      "last_updated": "2026-04-23",
+      "modalities": {
+        "input": [
+          "text",
+          "image",
+          "pdf"
+        ],
+        "output": [
+          "text"
+        ]
+      },
+      "open_weights": false,
+      "cost": {
+        "input": 5,
+        "output": 30,
+        "cache_read": 0.5,
+        "context_over_200k": {
+          "input": 10,
+          "output": 45,
+          "cache_read": 1
+        }
+      },
+      "limit": {
+        "context": 1050000,
+        "input": 920000,
+        "output": 130000
+      },
+      "experimental": {
+        "modes": {
+          "fast": {
+            "cost": {
+              "input": 12.5,
+              "output": 75,
+              "cache_read": 1.25
+            },
+            "provider": {
+              "body": {
+                "service_tier": "priority"
+              }
+            }
+          }
+        }
+      }
+    },
     "gpt-5-mini": {
       "id": "gpt-5-mini",
       "name": "GPT-5 Mini",

data/lib/llm/context.rb

@@ -295,7 +295,6 @@ module LLM
     #  ractor work, in that order.
     # @return [Array<LLM::Function::Return>]
     def wait(strategy)
-      stream = @params[:stream]
       if LLM::Stream === stream && !stream.queue.empty?
         @queue = stream.queue
         @queue.wait(strategy)
@@ -306,6 +305,7 @@ module LLM
       end
     ensure
       @queue = nil
+      @stream = nil
     end
 
     ##
@@ -461,6 +461,7 @@ module LLM
 
     def bind!(stream, model, tools)
       return unless LLM::Stream === stream
+      @stream = stream
       stream.extra[:ctx] = self
       stream.extra[:tracer] = tracer
       stream.extra[:model] = model
@@ -469,10 +470,13 @@ module LLM
 
     def queue
       return @queue if @queue
-      stream = @params[:stream]
       stream.queue if LLM::Stream === stream
     end
 
+    def stream
+      @stream || @params[:stream]
+    end
+
     def load_skills(skills)
       [*skills].map { LLM::Skill.load(_1).to_tool(self) }
     end

data/lib/llm/mcp.rb

@@ -103,6 +103,21 @@ class LLM::MCP
     nil
   end
 
+  ##
+  # Starts the MCP client for the duration of a block and then stops it.
+  # @yield Runs with the MCP client started
+  # @raise [LocalJumpError]
+  #  When called without a block
+  # @raise [StandardError]
+  #  Propagates errors raised by {#start}, the block itself, or {#stop}
+  # @return [void]
+  def run
+    start
+    yield
+  ensure
+    stop
+  end
+
   ##
   # Configures an HTTP MCP transport to use a persistent connection pool
   # via the optional dependency [Net::HTTP::Persistent](https://github.com/drbrain/net-http-persistent)

data/lib/llm/message.rb

@@ -33,11 +33,15 @@ module LLM
     # Returns a Hash representation of the message.
     # @return [Hash]
     def to_h
-      {role:, content:, reasoning_content:,
-       compaction: extra.compaction,
-       tools: extra.tool_calls,
-       usage:,
-       original_tool_calls: extra.original_tool_calls}.compact
+      {
+        role:,
+        content:,
+        reasoning_content:,
+        compaction: extra.compaction,
+        tools: extra.tool_calls&.map { LLM::Object === _1 ? _1.to_h : _1 },
+        usage:,
+        original_tool_calls: extra.original_tool_calls
+      }.compact.then { preserve_nil_content(_1) }
     end
 
     ##
@@ -208,6 +212,11 @@ module LLM
 
     private
 
+    def preserve_nil_content(hash)
+      hash[:content] = content if content.nil?
+      hash
+    end
+
     def tool_calls
       @tool_calls ||= LLM::Object.from(extra.tool_calls || [])
     end

data/lib/llm/providers/deepseek/request_adapter/completion.rb

@@ -19,7 +19,7 @@ module LLM::DeepSeek::RequestAdapter
         if Hash === message
           {role: message[:role], content: adapt_content(message[:content])}
         elsif message.tool_call?
-          {role: message.role, content: nil, tool_calls: message.extra[:original_tool_calls]}
+          wrap(content: nil, tool_calls: message.extra[:original_tool_calls])
         else
           adapt_message
         end
@@ -30,25 +30,34 @@ module LLM::DeepSeek::RequestAdapter
 
     def adapt_content(content)
       case content
+      when LLM::Object
+        adapt_object(content)
       when String
-        content.to_s
+        [{type: :text, text: content.to_s}]
       when LLM::Message
         adapt_content(content.content)
       when LLM::Function::Return
         throw(:abort, {role: "tool", tool_call_id: content.id, content: LLM.json.dump(content.value)})
-      when LLM::Object
-        prompt_error!(content)
       else
         prompt_error!(content)
       end
     end
 
+    def adapt_object(object)
+      case object.kind
+      when :image_url, :local_file, :remote_file
+        prompt_error!(object)
+      else
+        prompt_error!(object)
+      end
+    end
+
     def adapt_message
       case content
       when Array
         adapt_array
       else
-        {role: message.role, content: adapt_content(content)}
+        wrap(content: adapt_content(content))
       end
     end
 
@@ -58,13 +67,13 @@ module LLM::DeepSeek::RequestAdapter
       elsif returns.any?
         returns.map { {role: "tool", tool_call_id: _1.id, content: LLM.json.dump(_1.value)} }
       else
-        {role: message.role, content: content.flat_map { adapt_content(_1) }}
+        wrap(content: content.flat_map { adapt_content(_1) })
       end
     end
 
     def prompt_error!(object)
       if LLM::Object === object
-        raise LLM::PromptError, "The given LLM::Object with kind '#{content.kind}' is not " \
+        raise LLM::PromptError, "The given LLM::Object with kind '#{object.kind}' is not " \
                                 "supported by the DeepSeek API"
       else
         raise LLM::PromptError, "The given object (an instance of #{object.class}) " \
@@ -72,8 +81,22 @@ module LLM::DeepSeek::RequestAdapter
       end
     end
 
+    def wrap(content:, tool_calls: nil)
+      {
+        role: message.role,
+        content:,
+        tool_calls: tool_calls&.map { LLM::Object === _1 ? _1.to_h : _1 },
+        reasoning_content: message.reasoning_content
+      }.compact.then { preserve_nil_content(_1) }
+    end
+
     def message = @message
     def content = message.content
     def returns = content.grep(LLM::Function::Return)
+
+    def preserve_nil_content(hash)
+      hash[:content] = content if content.nil?
+      hash
+    end
   end
 end

data/lib/llm/providers/deepseek.rb

@@ -15,7 +15,7 @@ module LLM
   #
   #   llm = LLM.deepseek(key: ENV["KEY"])
   #   ctx = LLM::Context.new(llm)
-  #   ctx.talk ["Tell me about this photo", ctx.local_file("/images/photo.png")]
+  #   ctx.talk "Hello"
   #   ctx.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class DeepSeek < OpenAI
     require_relative "deepseek/request_adapter"
@@ -73,10 +73,10 @@ module LLM
 
     ##
     # Returns the default model for chat completions
-    # @see https://api-docs.deepseek.com/quick_start/pricing deepseek-chat
+    # @see https://api-docs.deepseek.com/quick_start/pricing deepseek-v4-flash
     # @return [String]
     def default_model
-      "deepseek-chat"
+      "deepseek-v4-flash"
     end
   end
 end

data/lib/llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LLM
-  VERSION = "5.1.0"
+  VERSION = "5.2.1"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llm.rb
 version: !ruby/object:Gem::Version
-  version: 5.1.0
+  version: 5.2.1
 platform: ruby
 authors:
 - Antar Azri