Add first-class self-hosted LLM support

- Add --base-url, --api-key, --adapter global CLI flags with env var fallbacks
- Make API key optional for local servers (localhost, custom domains)
- Add explicit adapter routing (openai/anthropic/ollama) via config or CLI
- Add `diffscope doctor` command for endpoint diagnostics and model recommendations
- Wire prompt optimization for local models with smaller context windows
- Consolidate ModelConfig construction via Config::to_model_config()
- Add release profile (LTO, strip), Docker dependency caching, Ollama healthcheck
- Add self-hosted documentation section and example configs

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: haasonsaas

Cargo.toml

@@ -35,6 +35,11 @@ shell-words = "1.1"
 tempfile = "3.8"
 mockito = "1.2"
 
+[profile.release]
+lto = "thin"
+codegen-units = 1
+strip = true
+
 [[bin]]
 name = "diffscope"
 path = "src/main.rs"

Dockerfile

@@ -4,19 +4,23 @@ FROM rust:alpine AS builder
 RUN apk add --no-cache musl-dev
 
 WORKDIR /app
+
+# Cache dependencies
 COPY Cargo.toml Cargo.lock ./
-COPY src ./src
+RUN mkdir src && echo 'fn main() {}' > src/main.rs && \
+    cargo build --release && \
+    rm -rf src
 
-# Build for the native architecture
-RUN cargo build --release
-RUN strip target/release/diffscope
+# Build actual binary
+COPY src ./src
+RUN touch src/main.rs && cargo build --release
 
 # Runtime stage
 FROM alpine:3.19
 
-RUN apk add --no-cache ca-certificates
+RUN apk add --no-cache ca-certificates git
 
 COPY --from=builder /app/target/release/diffscope /usr/local/bin/diffscope
 
 ENTRYPOINT ["diffscope"]
-CMD ["--help"]
+CMD ["--help"]

README.md

@@ -191,6 +191,63 @@ export OPENAI_BASE_URL=https://api.custom.com/v1
 git diff | diffscope review --model custom-model
 ```
 
+### Self-Hosted / Local Models
+
+Run DiffScope against a local LLM with zero cloud dependencies. No API key required.
+
+#### Ollama (Recommended)
+```bash
+# Install Ollama and pull a code model
+ollama pull codellama
+
+# Review code with local model
+git diff | diffscope review --base-url http://localhost:11434 --model ollama:codellama
+
+# Or use a config file (see examples/selfhosted-ollama.yml)
+cp examples/selfhosted-ollama.yml .diffscope.yml
+git diff | diffscope review
+```
+
+#### vLLM / LM Studio / OpenAI-Compatible Servers
+```bash
+# Point to any OpenAI-compatible endpoint
+git diff | diffscope review \
+  --base-url http://localhost:8000/v1 \
+  --adapter openai \
+  --model deepseek-coder-6.7b
+
+# See examples/selfhosted-vllm.yml for a ready-made config
+```
+
+#### Docker Compose (Ollama + DiffScope)
+```bash
+# Start Ollama and DiffScope together
+docker compose up diffscope-local
+
+# Pull a model first
+docker compose exec ollama ollama pull codellama
+```
+
+#### Check Your Setup
+```bash
+# Verify endpoint reachability, models, and recommendations
+diffscope doctor
+diffscope doctor --base-url http://localhost:11434
+```
+
+#### Environment Variables
+| Variable | Description |
+|----------|-------------|
+| `DIFFSCOPE_BASE_URL` | LLM API base URL (also accepts `OPENAI_BASE_URL`) |
+| `DIFFSCOPE_API_KEY` | API key for the LLM endpoint |
+
+#### CLI Flags
+| Flag | Description |
+|------|-------------|
+| `--base-url` | LLM API base URL |
+| `--api-key` | API key (optional for local servers) |
+| `--adapter` | Force adapter: `openai`, `anthropic`, or `ollama` |
+
 ### Supported Models
 
 **OpenAI**: gpt-4o, gpt-4-turbo, gpt-3.5-turbo

docker-compose.yml

@@ -11,25 +11,33 @@ services:
       - .:/workspace
     working_dir: /workspace
     command: review --diff /workspace/example.diff
-    
-  diffscope-ollama:
+
+  diffscope-local:
     build: .
     image: diffscope:latest
     depends_on:
-      - ollama
+      ollama:
+        condition: service_healthy
     environment:
-      - DIFFSCOPE_MODEL=ollama:codellama
+      - DIFFSCOPE_BASE_URL=http://ollama:11434
+      - DIFFSCOPE_MODEL=${DIFFSCOPE_MODEL:-ollama:codellama}
     volumes:
       - .:/workspace
     working_dir: /workspace
-    command: review --diff /workspace/example.diff --model ollama:codellama
-    
+    command: review --diff /workspace/example.diff --base-url http://ollama:11434 --model ollama:codellama
+
   ollama:
     image: ollama/ollama:latest
     ports:
       - "11434:11434"
     volumes:
       - ollama_data:/root/.ollama
-      
+    healthcheck:
+      test: ["CMD-SHELL", "curl -sf http://localhost:11434/api/tags || exit 1"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 30s
+
 volumes:
-  ollama_data:
+  ollama_data:

examples/selfhosted-ollama.yml

@@ -0,0 +1,19 @@
+# DiffScope config for self-hosted Ollama
+# Usage: cp examples/selfhosted-ollama.yml .diffscope.yml
+#        ollama pull codellama
+#        git diff | diffscope review
+
+model: ollama:codellama
+base_url: http://localhost:11434
+context_window: 8192
+temperature: 0.1
+max_tokens: 4096
+strictness: 2
+
+# No API key needed for local Ollama
+
+# Reduce context to fit smaller model windows
+max_context_chars: 8000
+max_diff_chars: 16000
+context_max_chunks: 8
+context_budget_chars: 8000

examples/selfhosted-vllm.yml

@@ -0,0 +1,20 @@
+# DiffScope config for self-hosted vLLM / LM Studio / any OpenAI-compatible server
+# Usage: cp examples/selfhosted-vllm.yml .diffscope.yml
+#        vllm serve deepseek-ai/deepseek-coder-6.7b-instruct
+#        git diff | diffscope review
+
+model: deepseek-ai/deepseek-coder-6.7b-instruct
+adapter: openai
+base_url: http://localhost:8000/v1
+context_window: 16384
+temperature: 0.1
+max_tokens: 4096
+strictness: 2
+
+# No API key needed for local servers
+
+# Adjust context for model capacity
+max_context_chars: 12000
+max_diff_chars: 24000
+context_max_chunks: 12
+context_budget_chars: 12000

src/adapters/anthropic.rs

@@ -50,17 +50,20 @@ struct AnthropicUsage {
 
 impl AnthropicAdapter {
     pub fn new(config: ModelConfig) -> Result<Self> {
-        let api_key = config.api_key.clone()
-            .or_else(|| std::env::var("ANTHROPIC_API_KEY").ok())
-            .context("Anthropic API key not found. Set ANTHROPIC_API_KEY environment variable or provide in config")?;
-
         let base_url = config
             .base_url
             .clone()
             .unwrap_or_else(|| "https://api.anthropic.com/v1".to_string());
 
+        let is_local = is_local_endpoint(&base_url);
+
+        let api_key = config.api_key.clone()
+            .or_else(|| std::env::var("ANTHROPIC_API_KEY").ok())
+            .or_else(|| if is_local { Some(String::new()) } else { None })
+            .context("Anthropic API key not found. Set ANTHROPIC_API_KEY environment variable or provide in config")?;
+
         let client = Client::builder()
-            .timeout(std::time::Duration::from_secs(60))
+            .timeout(std::time::Duration::from_secs(if is_local { 300 } else { 60 }))
             .build()?;
 
         Ok(Self {
@@ -176,3 +179,9 @@ impl LLMAdapter for AnthropicAdapter {
 fn is_retryable_status(status: StatusCode) -> bool {
     status == StatusCode::TOO_MANY_REQUESTS || status.is_server_error()
 }
+
+fn is_local_endpoint(url: &str) -> bool {
+    url.contains("localhost") || url.contains("127.0.0.1") || url.contains("0.0.0.0")
+        || url.contains("[::1]")
+        || (!url.contains("openai.com") && !url.contains("anthropic.com"))
+}

src/adapters/llm.rs

@@ -10,6 +10,8 @@ pub struct ModelConfig {
     pub temperature: f32,
     pub max_tokens: usize,
     pub openai_use_responses: Option<bool>,
+    #[serde(default)]
+    pub adapter_override: Option<String>,
 }
 
 impl Default for ModelConfig {
@@ -21,6 +23,7 @@ impl Default for ModelConfig {
             temperature: 0.2,
             max_tokens: 4000,
             openai_use_responses: None,
+            adapter_override: None,
         }
     }
 }
@@ -54,6 +57,16 @@ pub trait LLMAdapter: Send + Sync {
 }
 
 pub fn create_adapter(config: &ModelConfig) -> Result<Box<dyn LLMAdapter>> {
+    // Explicit adapter override takes priority
+    if let Some(ref adapter) = config.adapter_override {
+        return match adapter.as_str() {
+            "anthropic" => Ok(Box::new(crate::adapters::AnthropicAdapter::new(config.clone())?)),
+            "ollama" => Ok(Box::new(crate::adapters::OllamaAdapter::new(config.clone())?)),
+            _ => Ok(Box::new(crate::adapters::OpenAIAdapter::new(config.clone())?)),
+        };
+    }
+
+    // Model-name heuristic
     match config.model_name.as_str() {
         // Anthropic Claude models (all versions)
         name if name.starts_with("claude-") => Ok(Box::new(

src/adapters/openai.rs

@@ -87,17 +87,20 @@ struct OpenAIResponsesUsage {
 
 impl OpenAIAdapter {
     pub fn new(config: ModelConfig) -> Result<Self> {
-        let api_key = config.api_key.clone()
-            .or_else(|| std::env::var("OPENAI_API_KEY").ok())
-            .context("OpenAI API key not found. Set OPENAI_API_KEY environment variable or provide in config")?;
-
         let base_url = config
             .base_url
             .clone()
             .unwrap_or_else(|| "https://api.openai.com/v1".to_string());
 
+        let is_local = is_local_endpoint(&base_url);
+
+        let api_key = config.api_key.clone()
+            .or_else(|| std::env::var("OPENAI_API_KEY").ok())
+            .or_else(|| if is_local { Some(String::new()) } else { None })
+            .context("OpenAI API key not found. Set OPENAI_API_KEY environment variable or provide in config")?;
+
         let client = Client::builder()
-            .timeout(std::time::Duration::from_secs(60))
+            .timeout(std::time::Duration::from_secs(if is_local { 300 } else { 60 }))
             .build()?;
 
         Ok(Self {
@@ -164,6 +167,12 @@ fn is_retryable_status(status: StatusCode) -> bool {
     status == StatusCode::TOO_MANY_REQUESTS || status.is_server_error()
 }
 
+fn is_local_endpoint(url: &str) -> bool {
+    url.contains("localhost") || url.contains("127.0.0.1") || url.contains("0.0.0.0")
+        || url.contains("[::1]")
+        || (!url.contains("openai.com") && !url.contains("anthropic.com"))
+}
+
 fn should_use_responses_api(config: &ModelConfig) -> bool {
     if let Some(flag) = config.openai_use_responses {
         return flag;