dspy-gemini 1.0.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: b2b48f3533914c6239c1c655e229bac9479b3cc072cb4ae2916d76aa2b168386
+  data.tar.gz: b2a95b2f9710eb5c5ca0a944afefdc8e3552c495b76194c22a8f501d05f257b0
+SHA512:
+  metadata.gz: 3cbc0bc6a1482ed5168042e310e6135bcec8c17e533ee0fe86b3db70ca2aeff8efb8d5aec1fc05c27d70a2e47292d106802057b18cf374959ca37e9f545d33d3
+  data.tar.gz: 5bda92626cd9618697ca9f0cf993079258ae6add0df7f5bea7682d6c3e7031ae6a01fd6ba80d410b005d4086e21f010cc8dd39978960c38491f5ccebb4a85c9d

data/LICENSE

@@ -0,0 +1,45 @@
+MIT License
+
+Copyright (c) 2025 Vicente Services SL
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+This project is a Ruby port of the original Python [DSPy library](https://github.com/stanfordnlp/dspy), which is licensed under the MIT License:
+
+MIT License
+
+Copyright (c) 2023 Stanford Future Data Systems
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

data/README.md

@@ -0,0 +1,292 @@
+# DSPy.rb
+
+[![Gem Version](https://img.shields.io/gem/v/dspy)](https://rubygems.org/gems/dspy)
+[![Total Downloads](https://img.shields.io/gem/dt/dspy)](https://rubygems.org/gems/dspy)
+[![Build Status](https://img.shields.io/github/actions/workflow/status/vicentereig/dspy.rb/ruby.yml?branch=main&label=build)](https://github.com/vicentereig/dspy.rb/actions/workflows/ruby.yml)
+[![Documentation](https://img.shields.io/badge/docs-vicentereig.github.io%2Fdspy.rb-blue)](https://vicentereig.github.io/dspy.rb/)
+[![Discord](https://img.shields.io/discord/1161519468141355160?label=discord&logo=discord&logoColor=white)](https://discord.gg/zWBhrMqn)
+
+> [!NOTE]
+> The core Prompt Engineering Framework is production-ready with
+> comprehensive documentation. I am focusing now on educational content on systematic Prompt Optimization and Context Engineering.
+> Your feedback is invaluable. if you encounter issues, please open an [issue](https://github.com/vicentereig/dspy.rb/issues). If you have suggestions, open a [new thread](https://github.com/vicentereig/dspy.rb/discussions).  
+> 
+> If you want to contribute, feel free to reach out to me to coordinate efforts: hey at vicente.services
+>
+> And, yes, this is 100% a legit project. :) 
+
+
+**Build reliable LLM applications in idiomatic Ruby using composable, type-safe modules.**
+
+DSPy.rb is the Ruby-first surgical port of Stanford's [DSPy framework](https://github.com/stanfordnlp/dspy). It delivers structured LLM programming, prompt engineering, and context engineering in the language we love. Instead of wrestling with brittle prompt strings, you define typed signatures in idiomatic Ruby and compose workflows and agents that actually behave.
+
+**Prompts are just functions.** Traditional prompting is like writing code with string concatenation: it works until it doesn't. DSPy.rb brings you the programming approach pioneered by [dspy.ai](https://dspy.ai/): define modular signatures and let the framework deal with the messy bits.
+
+While we implement the same signatures, predictors, and optimization algorithms as the original library, DSPy.rb leans hard into Ruby conventions with Sorbet-based typing, ReAct loops, and production-ready integrations like non-blocking OpenTelemetry instrumentation.
+
+**What you get?** Ruby LLM applications that scale and don't break when you sneeze.
+
+Check the [examples](examples/) and take them for a spin!
+
+## Your First DSPy Program
+### Installation
+
+Add to your Gemfile:
+
+```ruby
+gem 'dspy'
+```
+
+and
+
+```bash
+bundle install
+```
+
+### Your First Reliable Predictor
+
+```ruby
+require 'dspy'
+
+# Configure DSPy globally to use your fave LLM (you can override per predictor).
+DSPy.configure do |c|
+  c.lm = DSPy::LM.new('openai/gpt-4o-mini',
+                      api_key: ENV['OPENAI_API_KEY'],
+                      structured_outputs: true)  # Enable OpenAI's native JSON mode
+end
+
+# Define a signature for sentiment classification - instead of writing a full prompt!
+class Classify < DSPy::Signature
+  description "Classify sentiment of a given sentence." # sets the goal of the underlying prompt
+
+  class Sentiment < T::Enum
+    enums do
+      Positive = new('positive')
+      Negative = new('negative')
+      Neutral = new('neutral')
+    end
+  end
+  
+  # Structured Inputs: makes sure you are sending only valid prompt inputs to your model
+  input do
+    const :sentence, String, description: 'The sentence to analyze'
+  end
+
+  # Structured Outputs: your predictor will validate the output of the model too.
+  output do
+    const :sentiment, Sentiment, description: 'The sentiment of the sentence'
+    const :confidence, Float, description: 'A number between 0.0 and 1.0'
+  end
+end
+
+# Wire it to the simplest prompting technique: a prediction loop.
+classify = DSPy::Predict.new(Classify)
+# it may raise an error if you mess the inputs or your LLM messes the outputs.
+result = classify.call(sentence: "This book was super fun to read!")
+
+puts result.sentiment    # => #<Sentiment::Positive>  
+puts result.confidence   # => 0.85
+```
+
+Save this as `examples/first_predictor.rb` and run it with:
+
+```bash
+bundle exec ruby examples/first_predictor.rb
+```
+
+### Sibling Gems
+
+DSPy.rb ships multiple gems from this monorepo so you can opt into features with heavier dependency trees (e.g., datasets pull in Polars/Arrow, MIPROv2 requires `numo-*` BLAS bindings) only when you need them. Add these alongside `dspy`:
+
+| Gem | Description | Status |
+| --- | --- | --- |
+| `dspy-schema` | Exposes `DSPy::TypeSystem::SorbetJsonSchema` for downstream reuse. (Still required by the core `dspy` gem; extraction lets other projects depend on it directly.) | **Stable** (v1.0.0) |
+| `dspy-code_act` | Think-Code-Observe agents that synthesize and execute Ruby safely. (Add the gem or set `DSPY_WITH_CODE_ACT=1` before requiring `dspy/code_act`.) | **Stable** (v1.0.0) |
+| `dspy-datasets` | Dataset helpers plus Parquet/Polars tooling for richer evaluation corpora. (Toggle via `DSPY_WITH_DATASETS`.) | **Stable** (v1.0.0) |
+| `dspy-evals` | High-throughput evaluation harness with metrics, callbacks, and regression fixtures. (Toggle via `DSPY_WITH_EVALS`.) | **Stable** (v1.0.0) |
+| `dspy-miprov2` | Bayesian optimization + Gaussian Process backend for the MIPROv2 teleprompter. (Install or export `DSPY_WITH_MIPROV2=1` before requiring the teleprompter.) | **Stable** (v1.0.0) |
+| `dspy-gepa` | `DSPy::Teleprompt::GEPA`, reflection loops, experiment tracking, telemetry adapters. (Install or set `DSPY_WITH_GEPA=1`.) | **Stable** (v1.0.0) |
+| `gepa` | GEPA optimizer core (Pareto engine, telemetry, reflective proposer). | **Stable** (v1.0.0) |
+| `dspy-o11y` | Core observability APIs: `DSPy::Observability`, async span processor, observation types. (Install or set `DSPY_WITH_O11Y=1`.) | **Stable** (v1.0.0) |
+| `dspy-o11y-langfuse` | Auto-configures DSPy observability to stream spans to Langfuse via OTLP. (Install or set `DSPY_WITH_O11Y_LANGFUSE=1`.) | **Stable** (v1.0.0) |
+| `dspy-deep_search` | Production DeepSearch loop with Exa-backed search/read, token budgeting, and instrumentation (Issue #163). | **Stable** (v1.0.0) |
+| `dspy-deep_research` | Planner/QA orchestration atop DeepSearch plus the memory supervisor used by the CLI example. | **Stable** (v1.0.0) |
+
+Set the matching `DSPY_WITH_*` environment variables (see `Gemfile`) to include or exclude each sibling gem when running Bundler locally (for example `DSPY_WITH_GEPA=1` or `DSPY_WITH_O11Y_LANGFUSE=1`). Refer to `adr/013-dependency-tree.md` for the full dependency map and roadmap.
+### Access to 200+ Models Across 5 Providers
+
+DSPy.rb provides unified access to major LLM providers with provider-specific optimizations:
+
+```ruby
+# OpenAI (GPT-4, GPT-4o, GPT-4o-mini, GPT-5, etc.)
+DSPy.configure do |c|
+  c.lm = DSPy::LM.new('openai/gpt-4o-mini',
+                      api_key: ENV['OPENAI_API_KEY'],
+                      structured_outputs: true)  # Native JSON mode
+end
+
+# Google Gemini (Gemini 1.5 Pro, Flash, Gemini 2.0, etc.)
+DSPy.configure do |c|
+  c.lm = DSPy::LM.new('gemini/gemini-2.5-flash',
+                      api_key: ENV['GEMINI_API_KEY'],
+                      structured_outputs: true)  # Native structured outputs
+end
+
+# Anthropic Claude (Claude 3.5, Claude 4, etc.)
+DSPy.configure do |c|
+  c.lm = DSPy::LM.new('anthropic/claude-sonnet-4-5-20250929',
+                      api_key: ENV['ANTHROPIC_API_KEY'],
+                      structured_outputs: true)  # Tool-based extraction (default)
+end
+
+# Ollama - Run any local model (Llama, Mistral, Gemma, etc.)
+DSPy.configure do |c|
+  c.lm = DSPy::LM.new('ollama/llama3.2')  # Free, runs locally, no API key needed
+end
+
+# OpenRouter - Access to 200+ models from multiple providers
+DSPy.configure do |c|
+  c.lm = DSPy::LM.new('openrouter/deepseek/deepseek-chat-v3.1:free',
+                      api_key: ENV['OPENROUTER_API_KEY'])
+end
+```
+
+## What You Get
+
+**Developer Experience:** Official clients, multimodal coverage, and observability baked in.
+<details>
+<summary>Expand for everything included</summary>
+
+- LLM provider support using official Ruby clients:
+  - [OpenAI Ruby](https://github.com/openai/openai-ruby) with vision model support
+  - [Anthropic Ruby SDK](https://github.com/anthropics/anthropic-sdk-ruby) with multimodal capabilities
+  - [Google Gemini API](https://ai.google.dev/) with native structured outputs
+  - [Ollama](https://ollama.com/) via OpenAI compatibility layer for local models
+- **Multimodal Support** - Complete image analysis with DSPy::Image, type-safe bounding boxes, vision-capable models
+- Runtime type checking with [Sorbet](https://sorbet.org/) including T::Enum and union types
+- Type-safe tool definitions for ReAct agents
+- Comprehensive instrumentation and observability
+</details>
+
+**Core Building Blocks:** Predictors, agents, and pipelines wired through type-safe signatures.
+<details>
+<summary>Expand for everything included</summary>
+
+- **Signatures** - Define input/output schemas using Sorbet types with T::Enum and union type support
+- **Predict** - LLM completion with structured data extraction and multimodal support
+- **Chain of Thought** - Step-by-step reasoning for complex problems with automatic prompt optimization
+- **ReAct** - Tool-using agents with type-safe tool definitions and error recovery
+- **Module Composition** - Combine multiple LLM calls into production-ready workflows
+</details>
+
+**Optimization & Evaluation:** Treat prompt optimization like a real ML workflow.
+<details>
+<summary>Expand for everything included</summary>
+
+- **Prompt Objects** - Manipulate prompts as first-class objects instead of strings
+- **Typed Examples** - Type-safe training data with automatic validation
+- **Evaluation Framework** - Advanced metrics beyond simple accuracy with error-resilient pipelines
+- **MIPROv2 Optimization** - Advanced Bayesian optimization with Gaussian Processes, multiple optimization strategies, auto-config presets, and storage persistence
+</details>
+
+**Production Features:** Hardened behaviors for teams shipping actual products.
+<details>
+<summary>Expand for everything included</summary>
+
+- **Reliable JSON Extraction** - Native structured outputs for OpenAI and Gemini, Anthropic tool-based extraction, and automatic strategy selection with fallback
+- **Type-Safe Configuration** - Strategy enums with automatic provider optimization (Strict/Compatible modes)
+- **Smart Retry Logic** - Progressive fallback with exponential backoff for handling transient failures
+- **Zero-Config Langfuse Integration** - Set env vars and get automatic OpenTelemetry traces in Langfuse
+- **Performance Caching** - Schema and capability caching for faster repeated operations
+- **File-based Storage** - Optimization result persistence with versioning
+- **Structured Logging** - JSON and key=value formats with span tracking
+</details>
+
+## Recent Achievements
+
+DSPy.rb has gone from experimental to production-ready in three fast releases.
+<details>
+<summary>Expand for the full changelog highlights</summary>
+
+### Foundation
+- ✅ **JSON Parsing Reliability** - Native OpenAI structured outputs with adaptive retry logic and schema-aware fallbacks
+- ✅ **Type-Safe Strategy Configuration** - Provider-optimized strategy selection and enum-backed optimizer presets
+- ✅ **Core Module System** - Predict, ChainOfThought, ReAct with type safety (add `dspy-code_act` for Think-Code-Observe agents)
+- ✅ **Production Observability** - OpenTelemetry, New Relic, and Langfuse integration
+- ✅ **Advanced Optimization** - MIPROv2 with Bayesian optimization, Gaussian Processes, and multi-mode search
+
+### Recent Advances
+- ✅ **MIPROv2 ADE Integrity (v0.29.1)** - Stratified train/val/test splits, honest precision accounting, and enum-driven `--auto` presets with integration coverage
+- ✅ **Instruction Deduplication (v0.29.1)** - Candidate generation now filters repeated programs so optimization logs highlight unique strategies
+- ✅ **GEPA Teleprompter (v0.29.0)** - Genetic-Pareto reflective prompt evolution with merge proposer scheduling, reflective mutation, and ADE demo parity
+- ✅ **Optimizer Utilities Parity (v0.29.0)** - Bootstrap strategies, dataset summaries, and Layer 3 utilities unlock multi-predictor programs on Ruby
+- ✅ **Observability Hardening (v0.29.0)** - OTLP exporter runs on a single-thread executor preventing frozen SSL contexts without blocking spans
+- ✅ **Documentation Refresh (v0.29.x)** - New GEPA guide plus ADE optimization docs covering presets, stratified splits, and error-handling defaults
+</details>
+
+**Current Focus Areas:** Closing the loop on production patterns and community adoption ahead of v1.0.
+<details>
+<summary>Expand for the roadmap</summary>
+
+### Production Readiness
+- 🚧 **Production Patterns** - Real-world usage validation and performance optimization
+- 🚧 **Ruby Ecosystem Integration** - Rails integration, Sidekiq compatibility, deployment patterns
+
+### Community & Adoption
+- 🚧 **Community Examples** - Real-world applications and case studies
+- 🚧 **Contributor Experience** - Making it easier to contribute and extend
+- 🚧 **Performance Benchmarks** - Comparative analysis vs other frameworks
+</details>
+
+**v1.0 Philosophy:** v1.0 lands after battle-testing, not checkbox bingo. The API is already stable; the milestone marks production confidence.
+
+
+## Documentation
+
+📖 **[Complete Documentation Website](https://vicentereig.github.io/dspy.rb/)**
+
+### LLM-Friendly Documentation
+
+For LLMs and AI assistants working with DSPy.rb:
+- **[llms.txt](https://vicentereig.github.io/dspy.rb/llms.txt)** - Concise reference optimized for LLMs
+- **[llms-full.txt](https://vicentereig.github.io/dspy.rb/llms-full.txt)** - Comprehensive API documentation
+
+### Getting Started
+- **[Installation & Setup](docs/src/getting-started/installation.md)** - Detailed installation and configuration
+- **[Quick Start Guide](docs/src/getting-started/quick-start.md)** - Your first DSPy programs
+- **[Core Concepts](docs/src/getting-started/core-concepts.md)** - Understanding signatures, predictors, and modules
+
+### Prompt Engineering
+- **[Signatures & Types](docs/src/core-concepts/signatures.md)** - Define typed interfaces for LLM operations
+- **[Predictors](docs/src/core-concepts/predictors.md)** - Predict, ChainOfThought, ReAct, and more
+- **[Modules & Pipelines](docs/src/core-concepts/modules.md)** - Compose complex multi-stage workflows
+- **[Multimodal Support](docs/src/core-concepts/multimodal.md)** - Image analysis with vision-capable models
+- **[Examples & Validation](docs/src/core-concepts/examples.md)** - Type-safe training data
+- **[Rich Types](docs/src/advanced/complex-types.md)** - Sorbet type integration with automatic coercion for structs, enums, and arrays
+- **[Composable Pipelines](docs/src/advanced/pipelines.md)** - Manual module composition patterns
+
+### Prompt Optimization
+- **[Evaluation Framework](docs/src/optimization/evaluation.md)** - Advanced metrics beyond simple accuracy
+- **[Prompt Optimization](docs/src/optimization/prompt-optimization.md)** - Manipulate prompts as objects
+- **[MIPROv2 Optimizer](docs/src/optimization/miprov2.md)** - Advanced Bayesian optimization with Gaussian Processes
+- **[GEPA Optimizer](docs/src/optimization/gepa.md)** *(beta)* - Reflective mutation with optional reflection LMs
+
+### Context Engineering
+- **[Tools](docs/src/core-concepts/toolsets.md)** - Tool wieldint agents.
+- **[Agentic Memory](docs/src/core-concepts/memory.md)** - Memory Tools & Agentic Loops
+- **[RAG Patterns](docs/src/advanced/rag.md)** - Manual RAG implementation with external services
+
+### Production Features
+- **[Observability](docs/src/production/observability.md)** - Zero-config Langfuse integration with a dedicated export worker that never blocks your LLMs
+- **[Storage System](docs/src/production/storage.md)** - Persistence and optimization result storage
+- **[Custom Metrics](docs/src/advanced/custom-metrics.md)** - Proc-based evaluation logic 
+
+
+
+
+
+
+
+
+## License
+This project is licensed under the MIT License.

data/lib/dspy/gemini/README.md

@@ -0,0 +1,37 @@
+# DSPy Gemini adapter gem
+
+`dspy-gemini` provides the Gemini adapter for DSPy.rb so we can rely on Google's API without bloating the core gem. Install it whenever you plan to call `gemini/*` model ids from DSPy.
+
+## When you need it
+- You call `DSPy::LM.new` with a provider of `gemini`.
+- You want structured outputs, multimodal prompts, or streaming responses backed by Gemini's Generative Language API.
+
+Projects that only target OpenAI, Anthropic, or other providers can skip this gem.
+
+## Installation
+Add it beside `dspy` and install dependencies:
+
+```ruby
+# Gemfile
+gem 'dspy'
+gem 'dspy-gemini'
+```
+
+```sh
+bundle install
+```
+
+The adapter enforces `gemini-ai ~> 4.3` at runtime (and raises if the dependency is missing or out of range).
+
+## Configuration
+- Set `ENV['GEMINI_API_KEY']`, or pass `api_key:` directly.
+
+## Basic usage
+
+```ruby
+require 'dspy'
+# No need to explicitly require 'dspy/gemini'
+
+lm = DSPy::LM.new('gemini/gemini-1.5-flash', api_key: ENV.fetch('GEMINI_API_KEY'))
+
+```

data/lib/dspy/gemini/guardrails.rb

@@ -0,0 +1,25 @@
+# frozen_string_literal: true
+
+require 'dspy/lm/errors'
+
+module DSPy
+  module Gemini
+    class Guardrails
+      SUPPORTED_GEMINI_VERSIONS = "~> 4.3".freeze
+
+      def self.ensure_gemini_installed!
+        require 'gemini-ai'
+
+        spec = Gem.loaded_specs["gemini-ai"]
+        unless spec && Gem::Requirement.new(SUPPORTED_GEMINI_VERSIONS).satisfied_by?(spec.version)
+          msg = <<~MSG
+            DSPY requires `gemini-ai` gem #{SUPPORTED_GEMINI_VERSIONS}.
+            Please Install or upgrade it with `bundle add gemini-ai --version "#{SUPPORTED_GEMINI_VERSIONS}"`.
+          MSG
+          raise DSPy::LM::UnsupportedVersionError, msg
+        end
+      end
+    end
+  end
+end
+

data/lib/dspy/gemini/lm/adapters/gemini_adapter.rb

@@ -0,0 +1,228 @@
+# frozen_string_literal: true
+
+require 'gemini-ai'
+require 'json'
+require 'dspy/lm/vision_models'
+require 'dspy/lm/adapter'
+
+require 'dspy/gemini/guardrails'
+DSPy::Gemini::Guardrails.ensure_gemini_installed!
+
+module DSPy
+  module Gemini
+    module LM
+      module Adapters
+        class GeminiAdapter < DSPy::LM::Adapter
+          def initialize(model:, api_key:, structured_outputs: false)
+            super(model: model, api_key: api_key)
+            validate_api_key!(api_key, 'gemini')
+
+            @structured_outputs_enabled = structured_outputs
+
+            # Disable streaming for VCR tests since SSE responses don't record properly
+            # But keep streaming enabled for SSEVCR tests (SSE-specific cassettes)
+            @use_streaming = true
+            begin
+              vcr_active = defined?(VCR) && VCR.current_cassette
+              ssevcr_active = defined?(SSEVCR) && SSEVCR.turned_on?
+
+              # Only disable streaming if regular VCR is active but SSEVCR is not
+              @use_streaming = false if vcr_active && !ssevcr_active
+            rescue
+              # If VCR/SSEVCR is not available or any error occurs, use streaming
+              @use_streaming = true
+            end
+
+            @client = ::Gemini.new(
+              credentials: {
+                service: 'generative-language-api',
+                api_key: api_key,
+                version: 'v1beta'  # Use beta API version for structured outputs support
+              },
+              options: { 
+                model: model,
+                server_sent_events: @use_streaming
+              }
+            )
+          end
+
+          def chat(messages:, signature: nil, **extra_params, &block)
+            normalized_messages = normalize_messages(messages)
+
+            # Validate vision support if images are present
+            if contains_images?(normalized_messages)
+              DSPy::LM::VisionModels.validate_vision_support!('gemini', model)
+              # Convert messages to Gemini format with proper image handling
+              normalized_messages = format_multimodal_messages(normalized_messages)
+            end
+
+            # Convert DSPy message format to Gemini format
+            gemini_messages = convert_messages_to_gemini_format(normalized_messages)
+
+            request_params = {
+              contents: gemini_messages
+            }.merge(extra_params)
+
+            begin
+              content = ""
+              final_response_data = nil
+
+              # Check if we're using streaming or not
+              if @use_streaming
+                # Streaming mode
+                @client.stream_generate_content(request_params) do |chunk|
+                  # Handle case where chunk might be a string (from SSE VCR)
+                  if chunk.is_a?(String)
+                    begin
+                      chunk = JSON.parse(chunk)
+                    rescue JSON::ParserError => e
+                      raise DSPy::LM::AdapterError, "Failed to parse Gemini streaming response: #{e.message}"
+                    end
+                  end
+
+                  # Extract content from chunks
+                  if chunk.dig('candidates', 0, 'content', 'parts')
+                    chunk_text = extract_text_from_parts(chunk.dig('candidates', 0, 'content', 'parts'))
+                    content += chunk_text
+
+                    # Call block only if provided (for real streaming)
+                    block.call(chunk) if block_given?
+                  end
+
+                  # Store final response data (usage, metadata) from last chunk
+                  if chunk['usageMetadata'] || chunk.dig('candidates', 0, 'finishReason')
+                    final_response_data = chunk
+                  end
+                end
+              else
+                # Non-streaming mode (for VCR tests)
+                response = @client.generate_content(request_params)
+
+                # Extract content from single response
+                if response.dig('candidates', 0, 'content', 'parts')
+                  content = extract_text_from_parts(response.dig('candidates', 0, 'content', 'parts'))
+                end
+
+                # Use response as final data
+                final_response_data = response
+              end
+
+              # Extract usage information from final chunk
+              usage_data = final_response_data&.dig('usageMetadata')
+              usage_struct = usage_data ? DSPy::LM::UsageFactory.create('gemini', usage_data) : nil
+
+              # Create metadata from final chunk
+              metadata = {
+                provider: 'gemini',
+                model: model,
+                finish_reason: final_response_data&.dig('candidates', 0, 'finishReason'),
+                safety_ratings: final_response_data&.dig('candidates', 0, 'safetyRatings'),
+                streaming: block_given?
+              }
+
+              # Create typed metadata
+              typed_metadata = DSPy::LM::ResponseMetadataFactory.create('gemini', metadata)
+
+              DSPy::LM::Response.new(
+                content: content,
+                usage: usage_struct,
+                metadata: typed_metadata
+              )
+            rescue => e
+              handle_gemini_error(e)
+            end
+          end
+
+          private
+
+          # Convert DSPy message format to Gemini format
+          def convert_messages_to_gemini_format(messages)
+            # Gemini expects contents array with role and parts
+            messages.map do |msg|
+              role = case msg[:role]
+                     when 'system'
+                       'user' # Gemini doesn't have explicit system role, merge with user
+                     when 'assistant'
+                       'model'
+                     else
+                       msg[:role]
+                     end
+
+              if msg[:content].is_a?(Array)
+                # Multimodal content
+                parts = msg[:content].map do |item|
+                  case item[:type]
+                  when 'text'
+                    { text: item[:text] }
+                  when 'image'
+                    item[:image].to_gemini_format
+                  else
+                    item
+                  end
+                end
+
+                { role: role, parts: parts }
+              else
+                # Text-only content
+                { role: role, parts: [{ text: msg[:content] }] }
+              end
+            end
+          end
+
+          # Extract text content from Gemini parts array
+          def extract_text_from_parts(parts)
+            return "" unless parts.is_a?(Array)
+
+            parts.map { |part| part['text'] }.compact.join
+          end
+
+          # Format multimodal messages for Gemini
+          def format_multimodal_messages(messages)
+            messages.map do |msg|
+              if msg[:content].is_a?(Array)
+                # Convert multimodal content to Gemini format
+                formatted_content = msg[:content].map do |item|
+                  case item[:type]
+                  when 'text'
+                    { type: 'text', text: item[:text] }
+                  when 'image'
+                    # Validate image compatibility before formatting
+                    item[:image].validate_for_provider!('gemini')
+                    item[:image].to_gemini_format
+                  else
+                    item
+                  end
+                end
+
+                {
+                  role: msg[:role],
+                  content: formatted_content
+                }
+              else
+                msg
+              end
+            end
+          end
+
+          # Handle Gemini-specific errors
+          def handle_gemini_error(error)
+            error_msg = error.message.to_s
+
+            if error_msg.include?('API_KEY') || error_msg.include?('status 400') || error_msg.include?('status 401') || error_msg.include?('status 403')
+              raise DSPy::LM::AdapterError, "Gemini authentication failed: #{error_msg}. Check your API key."
+            elsif error_msg.include?('RATE_LIMIT') || error_msg.downcase.include?('quota') || error_msg.include?('status 429')
+              raise DSPy::LM::AdapterError, "Gemini rate limit exceeded: #{error_msg}. Please wait and try again."
+            elsif error_msg.include?('SAFETY') || error_msg.include?('blocked')
+              raise DSPy::LM::AdapterError, "Gemini content was blocked by safety filters: #{error_msg}"
+            elsif error_msg.include?('image') || error_msg.include?('media')
+              raise DSPy::LM::AdapterError, "Gemini image processing failed: #{error_msg}. Ensure your image is a valid format and under size limits."
+            else
+              # Generic error handling
+              raise DSPy::LM::AdapterError, "Gemini adapter error: #{error_msg}"
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/dspy/gemini/lm/schema_converter.rb

@@ -0,0 +1,184 @@
+# frozen_string_literal: true
+
+require "sorbet-runtime"
+
+module DSPy
+  module Gemini
+    module LM
+      # Converts DSPy signatures to Gemini structured output format
+      class SchemaConverter
+        extend T::Sig
+
+        # Models that support structured outputs (JSON + Schema)
+        # Based on official Google documentation: https://ai.google.dev/gemini-api/docs/models/gemini
+        # Last updated: Oct 2025
+        # Note: Gemini 1.5 series deprecated Oct 2025
+        STRUCTURED_OUTPUT_MODELS = T.let([
+          # Gemini 2.0 series
+          "gemini-2.0-flash",
+          "gemini-2.0-flash-lite",
+          # Gemini 2.5 series (current)
+          "gemini-2.5-pro",
+          "gemini-2.5-flash",
+          "gemini-2.5-flash-lite",
+          "gemini-2.5-flash-image"
+        ].freeze, T::Array[String])
+
+        # Models that do not support structured outputs or are deprecated
+        UNSUPPORTED_MODELS = T.let([
+          # Legacy Gemini 1.0 series
+          "gemini-pro",
+          "gemini-1.0-pro-002",
+          "gemini-1.0-pro",
+          # Deprecated Gemini 1.5 series (removed Oct 2025)
+          "gemini-1.5-pro",
+          "gemini-1.5-pro-preview-0514",
+          "gemini-1.5-pro-preview-0409",
+          "gemini-1.5-flash",
+          "gemini-1.5-flash-8b"
+        ].freeze, T::Array[String])
+
+        sig { params(signature_class: T.class_of(DSPy::Signature)).returns(T::Hash[Symbol, T.untyped]) }
+        def self.to_gemini_format(signature_class)
+          # Get the output JSON schema from the signature class
+          output_schema = signature_class.output_json_schema
+
+          # Convert to Gemini format (OpenAPI 3.0 Schema subset - not related to OpenAI)
+          convert_dspy_schema_to_gemini(output_schema)
+        end
+
+        sig { params(model: String).returns(T::Boolean) }
+        def self.supports_structured_outputs?(model)
+          # Extract base model name without provider prefix
+          base_model = model.sub(/^gemini\//, "")
+
+          # Check if it's a supported model or a newer version
+          STRUCTURED_OUTPUT_MODELS.any? { |supported| base_model.start_with?(supported) }
+        end
+
+        sig { params(schema: T::Hash[Symbol, T.untyped]).returns(T::Array[String]) }
+        def self.validate_compatibility(schema)
+          issues = []
+
+          # Check for deeply nested objects (Gemini has depth limits)
+          depth = calculate_depth(schema)
+          if depth > 5
+            issues << "Schema depth (#{depth}) exceeds recommended limit of 5 levels"
+          end
+
+          issues
+        end
+
+        private
+
+        sig { params(dspy_schema: T::Hash[Symbol, T.untyped]).returns(T::Hash[Symbol, T.untyped]) }
+        def self.convert_dspy_schema_to_gemini(dspy_schema)
+          # For Gemini's responseJsonSchema, we need pure JSON Schema format
+          # Remove OpenAPI-specific fields like "$schema"
+          result = {
+            type: "object",
+            properties: {},
+            required: []
+          }
+
+          # Convert properties
+          properties = dspy_schema[:properties] || {}
+          properties.each do |prop_name, prop_schema|
+            result[:properties][prop_name] = convert_property_to_gemini(prop_schema)
+          end
+
+          # Set required fields
+          result[:required] = (dspy_schema[:required] || []).map(&:to_s)
+
+          result
+        end
+
+        sig { params(property_schema: T::Hash[Symbol, T.untyped]).returns(T::Hash[Symbol, T.untyped]) }
+        def self.convert_property_to_gemini(property_schema)
+          # Handle oneOf/anyOf schemas (union types) - Gemini supports these in responseJsonSchema
+          if property_schema[:oneOf]
+            return {
+              oneOf: property_schema[:oneOf].map { |schema| convert_property_to_gemini(schema) },
+              description: property_schema[:description]
+            }.compact
+          end
+
+          if property_schema[:anyOf]
+            return {
+              anyOf: property_schema[:anyOf].map { |schema| convert_property_to_gemini(schema) },
+              description: property_schema[:description]
+            }.compact
+          end
+
+          case property_schema[:type]
+          when "string"
+            result = { type: "string" }
+            # Gemini responseJsonSchema doesn't support const, so convert to single-value enum
+            # See: https://ai.google.dev/api/generate-content#FIELDS.response_json_schema
+            if property_schema[:const]
+              result[:enum] = [property_schema[:const]]
+            elsif property_schema[:enum]
+              result[:enum] = property_schema[:enum]
+            end
+            result
+          when "integer"
+            { type: "integer" }
+          when "number"
+            { type: "number" }
+          when "boolean"
+            { type: "boolean" }
+          when "array"
+            {
+              type: "array",
+              items: convert_property_to_gemini(property_schema[:items] || { type: "string" })
+            }
+          when "object"
+            result = { type: "object" }
+
+            if property_schema[:properties]
+              result[:properties] = {}
+              property_schema[:properties].each do |nested_prop, nested_schema|
+                result[:properties][nested_prop] = convert_property_to_gemini(nested_schema)
+              end
+
+              # Set required fields for nested objects
+              if property_schema[:required]
+                result[:required] = property_schema[:required].map(&:to_s)
+              end
+            end
+
+            result
+          else
+            # Default to string for unknown types
+            { type: "string" }
+          end
+        end
+
+        sig { params(schema: T::Hash[Symbol, T.untyped], current_depth: Integer).returns(Integer) }
+        def self.calculate_depth(schema, current_depth = 0)
+          return current_depth unless schema.is_a?(Hash)
+
+          max_depth = current_depth
+
+          # Check properties
+          if schema[:properties].is_a?(Hash)
+            schema[:properties].each_value do |prop|
+              if prop.is_a?(Hash)
+                prop_depth = calculate_depth(prop, current_depth + 1)
+                max_depth = [max_depth, prop_depth].max
+              end
+            end
+          end
+
+          # Check array items
+          if schema[:items].is_a?(Hash)
+            items_depth = calculate_depth(schema[:items], current_depth + 1)
+            max_depth = [max_depth, items_depth].max
+          end
+
+          max_depth
+        end
+      end
+    end
+  end
+end

data/lib/dspy/gemini/version.rb

@@ -0,0 +1,7 @@
+# frozen_string_literal: true
+
+module DSPy
+  module Gemini
+    VERSION = '1.0.0'
+  end
+end

data/lib/dspy/gemini.rb

@@ -0,0 +1,9 @@
+# frozen_string_literal: true
+
+require 'dspy/gemini/version'
+
+require 'dspy/gemini/guardrails'
+DSPy::Gemini::Guardrails.ensure_gemini_installed!
+
+require 'dspy/gemini/lm/adapters/gemini_adapter'
+require 'dspy/gemini/lm/schema_converter'

metadata

@@ -0,0 +1,78 @@
+--- !ruby/object:Gem::Specification
+name: dspy-gemini
+version: !ruby/object:Gem::Version
+  version: 1.0.0
+platform: ruby
+authors:
+- Vicente Reig Rincón de Arellano
+bindir: bin
+cert_chain: []
+date: 1980-01-02 00:00:00.000000000 Z
+dependencies:
+- !ruby/object:Gem::Dependency
+  name: dspy
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - '='
+      - !ruby/object:Gem::Version
+        version: 0.31.0
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - '='
+      - !ruby/object:Gem::Version
+        version: 0.31.0
+- !ruby/object:Gem::Dependency
+  name: gemini-ai
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+description: Provides the GeminiAdapter so Gemini-compatible providers can be added
+  to DSPy.rb projects independently of the core gem.
+email:
+- hey@vicente.services
+executables: []
+extensions: []
+extra_rdoc_files: []
+files:
+- LICENSE
+- README.md
+- lib/dspy/gemini.rb
+- lib/dspy/gemini/README.md
+- lib/dspy/gemini/guardrails.rb
+- lib/dspy/gemini/lm/adapters/gemini_adapter.rb
+- lib/dspy/gemini/lm/schema_converter.rb
+- lib/dspy/gemini/version.rb
+homepage: https://github.com/vicentereig/dspy.rb
+licenses:
+- MIT
+metadata:
+  github_repo: git@github.com:vicentereig/dspy.rb
+rdoc_options: []
+require_paths:
+- lib
+required_ruby_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: 3.3.0
+required_rubygems_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: '0'
+requirements: []
+rubygems_version: 3.6.9
+specification_version: 4
+summary: Gemini adapters for DSPy.rb.
+test_files: []