refactor: remove Ollama and add custom provider configuration (#546)

* feat(config): add ExecutionConfig for runtime behavior settings

Add ExecutionConfig struct for controlling runtime execution:
- max_agent_threads: Maximum concurrent agent threads
- max_tool_threads: Maximum concurrent tool executions
- command_timeout_secs: Timeout for shell commands
- http_timeout_secs: Timeout for HTTP requests
- streaming_enabled: Enable/disable streaming responses

This allows users to customize execution behavior via config files.

* feat(config): add custom providers configuration support

Add CustomProviderConfig and CustomModelConfig types for user-defined
LLM providers in config.toml:
- base_url: Provider API endpoint
- api_type: openai/anthropic/openai-compatible
- api_key_env: Environment variable for API key
- default_model: Default model for provider
- models: Available models with capabilities
- headers: Custom HTTP headers

Enables users to add their own providers without code changes.

* feat(cli): add runtime execution configuration arguments

Add CLI arguments for controlling execution behavior:
- --max-agent-threads: Maximum concurrent agent threads
- --max-tool-threads: Maximum concurrent tool executions
- --command-timeout: Timeout for shell commands in seconds
- --http-timeout: Timeout for HTTP requests in seconds
- --no-streaming: Disable streaming responses

These arguments override config file settings at runtime.

* refactor: remove cortex-ollama crate and all backend references

Remove Ollama integration from the backend:
- Delete cortex-ollama crate entirely
- Remove from workspace members and dependencies
- Remove Ollama model presets from cortex-common
- Remove OllamaEmbedder from cortex-engine
- Remove Ollama provider references from CLI utilities
- Remove Ollama options from app-server config
- Clean up related imports and references

Users can now add custom local providers via the new
custom providers configuration system.

* refactor(gui): remove Ollama provider from cortex-gui

Remove all Ollama provider references from the GUI:
- Delete OllamaProvider.ts implementation
- Remove Ollama from AIProvider enum (backend)
- Remove Ollama from LLMProviderType union type
- Remove Ollama from provider configuration objects
- Remove Ollama from model selectors and settings
- Remove Ollama from onboarding wizard
- Update package-lock.json dependencies

The custom providers system now allows users to configure
local inference endpoints without built-in Ollama support.

Author: echobt

Cargo.lock

@@ -80,9 +80,6 @@ members = [
     # CLI - LM Studio Integration
     "src/cortex-lmstudio",
 
-    # CLI - Ollama Integration
-    "src/cortex-ollama",
-    
     # CLI - TUI
     "src/cortex-core",
     "src/cortex-tui",
@@ -224,7 +221,6 @@ cortex-experimental = { path = "src/cortex-experimental" }
 cortex-linux-sandbox = { path = "src/cortex-linux-sandbox" }
 cortex-windows-sandbox = { path = "src/cortex-windows-sandbox" }
 cortex-lmstudio = { path = "src/cortex-lmstudio" }
-cortex-ollama = { path = "src/cortex-ollama" }
 cortex-skills = { path = "src/cortex-skills" }
 cortex-prompt-harness = { path = "src/cortex-prompt-harness" }
 

Cargo.toml

@@ -374,9 +374,6 @@ pub struct ProviderConfig {
     pub azure_api_key: Option<String>,
     /// Azure OpenAI deployment.
     pub azure_deployment: Option<String>,
-    /// Ollama base URL.
-    #[serde(default = "default_ollama_url")]
-    pub ollama_url: String,
     /// Default provider.
     #[serde(default = "default_provider")]
     pub default_provider: String,
@@ -385,10 +382,6 @@ pub struct ProviderConfig {
     pub default_model: String,
 }
 
-fn default_ollama_url() -> String {
-    "http://localhost:11434".to_string()
-}
-
 fn default_provider() -> String {
     "openai".to_string()
 }

src/cortex-app-server/src/config.rs

@@ -39,7 +39,7 @@ pub fn validate_model_name(model: &str) -> Result<String> {
             "mistral",
             "xai",
             "deepseek",
-            "ollama",
+            "groq",
         ];
         let provider = parts[0].to_lowercase();
         if !valid_providers.contains(&provider.as_str()) {

src/cortex-cli/src/agent_cmd/utils.rs

@@ -208,6 +208,26 @@ pub struct InteractiveArgs {
     #[arg(long = "search", default_value_t = false, help_heading = "Features")]
     pub web_search: bool,
 
+    /// Maximum number of concurrent agent threads
+    #[arg(long = "max-agent-threads", value_name = "N", help_heading = "Execution")]
+    pub max_agent_threads: Option<usize>,
+
+    /// Maximum number of concurrent tool executions
+    #[arg(long = "max-tool-threads", value_name = "N", help_heading = "Execution")]
+    pub max_tool_threads: Option<usize>,
+
+    /// Timeout for shell commands in seconds
+    #[arg(long = "command-timeout", value_name = "SECONDS", help_heading = "Execution")]
+    pub command_timeout: Option<u64>,
+
+    /// Timeout for HTTP requests in seconds
+    #[arg(long = "http-timeout", value_name = "SECONDS", help_heading = "Execution")]
+    pub http_timeout: Option<u64>,
+
+    /// Disable streaming responses
+    #[arg(long = "no-streaming", default_value_t = false, help_heading = "Execution")]
+    pub no_streaming: bool,
+
     /// Set log verbosity level (error, warn, info, debug, trace)
     #[arg(
         long = "log-level",

src/cortex-cli/src/cli/args.rs

@@ -411,35 +411,6 @@ fn get_available_models() -> Vec<ModelInfo> {
             input_cost_per_million: Some(0.55),
             output_cost_per_million: Some(2.19),
         },
-        // Local/OSS models (Ollama) - no API cost, run locally
-        ModelInfo {
-            id: "qwen2.5-coder:32b".to_string(),
-            name: "Qwen 2.5 Coder 32B".to_string(),
-            provider: "ollama".to_string(),
-            capabilities: ModelCapabilities {
-                vision: false,
-                tools: true,
-                parallel_tools: false, // Ollama models typically don't support parallel tools
-                streaming: true,
-                json_mode: true,
-            },
-            input_cost_per_million: None, // Local model, no API cost
-            output_cost_per_million: None,
-        },
-        ModelInfo {
-            id: "llama3.3:70b".to_string(),
-            name: "Llama 3.3 70B".to_string(),
-            provider: "ollama".to_string(),
-            capabilities: ModelCapabilities {
-                vision: false,
-                tools: true,
-                parallel_tools: false, // Ollama models typically don't support parallel tools
-                streaming: true,
-                json_mode: true,
-            },
-            input_cost_per_million: None, // Local model, no API cost
-            output_cost_per_million: None,
-        },
     ]
 }
 

src/cortex-cli/src/models_cmd.rs

@@ -286,11 +286,6 @@ fn get_model_pricing(model: &str) -> ModelPricing {
             input_per_million: 2.0,
             output_per_million: 10.0,
         },
-        // Local/OSS (free)
-        m if m.contains("ollama") || m.contains("qwen") || m.contains("llama3") => ModelPricing {
-            input_per_million: 0.0,
-            output_per_million: 0.0,
-        },
         // Default (conservative estimate)
         _ => ModelPricing {
             input_per_million: 3.0,
@@ -319,16 +314,16 @@ fn infer_provider(model: &str) -> String {
         "openai".to_string()
     } else if model_lower.contains("gemini") {
         "google".to_string()
-    } else if model_lower.contains("llama") && !model_lower.contains("ollama") {
+    } else if model_lower.contains("llama") {
         "groq".to_string()
     } else if model_lower.contains("mistral") || model_lower.contains("codestral") {
         "mistral".to_string()
     } else if model_lower.contains("deepseek") {
         "deepseek".to_string()
     } else if model_lower.contains("grok") {
         "xai".to_string()
-    } else if model_lower.contains("ollama") || model_lower.contains("qwen") {
-        "ollama".to_string()
+    } else if model_lower.contains("qwen") {
+        "deepseek".to_string() // Qwen models often available via DeepSeek
     } else {
         "unknown".to_string()
     }
@@ -735,9 +730,9 @@ mod tests {
         let cost = calculate_cost("claude-sonnet-4", 1_000_000, 1_000_000);
         assert!((cost - 18.0).abs() < 0.001);
 
-        // Free local model
-        let cost = calculate_cost("ollama:llama3", 1_000_000, 1_000_000);
-        assert_eq!(cost, 0.0);
+        // GPT-4o: $2.50/$10 per 1M
+        let cost = calculate_cost("gpt-4o", 1_000_000, 1_000_000);
+        assert!((cost - 12.5).abs() < 0.001);
     }
 
     #[test]

src/cortex-cli/src/stats_cmd.rs

@@ -13,7 +13,7 @@ pub const KNOWN_PROVIDERS: &[&str] = &[
     "mistral",
     "xai",
     "deepseek",
-    "ollama",
+    "groq",
     "lmstudio",
     "llamacpp",
     "vllm",

src/cortex-cli/src/utils/model.rs

@@ -303,7 +303,7 @@ pub fn validate_model_name(model: &str) -> Result<String> {
             "mistral",
             "xai",
             "deepseek",
-            "ollama",
+            "groq",
         ];
         let provider = parts[0].to_lowercase();
         if !valid_providers.contains(&provider.as_str()) {