refactor(runtime): improve core runtime internals

README.md

@@ -311,7 +311,7 @@ Read the full vision in [docs/spacedrive.md](docs/spacedrive.md).
 ### Prerequisites
 
 - **Rust** 1.85+ ([rustup](https://rustup.rs/))
-- An LLM API key from any supported provider (Anthropic, OpenAI, OpenRouter, Z.ai, Groq, Together, Fireworks, DeepSeek, xAI, Mistral, or OpenCode Zen)
+- An LLM API key from any supported provider (Anthropic, OpenAI, OpenRouter, Ollama Cloud, Z.ai, Groq, Together, Fireworks, DeepSeek, xAI, Mistral, or OpenCode Zen)
 
 ### Build and Run
 

build.rs

@@ -15,7 +15,9 @@ fn main() {
 
     // Skip if bun isn't installed or node_modules is missing (CI without frontend deps)
     if !interface_dir.join("node_modules").exists() {
-        eprintln!("cargo:warning=interface/node_modules not found, skipping frontend build. Run `bun install` in interface/");
+        eprintln!(
+            "cargo:warning=interface/node_modules not found, skipping frontend build. Run `bun install` in interface/"
+        );
         ensure_dist_dir();
         return;
     }
@@ -28,10 +30,14 @@ fn main() {
     match status {
         Ok(s) if s.success() => {}
         Ok(s) => {
-            eprintln!("cargo:warning=frontend build exited with {s}, the binary will serve a stale or empty UI");
+            eprintln!(
+                "cargo:warning=frontend build exited with {s}, the binary will serve a stale or empty UI"
+            );
         }
         Err(e) => {
-            eprintln!("cargo:warning=failed to run `bun run build`: {e}. Install bun to build the frontend.");
+            eprintln!(
+                "cargo:warning=failed to run `bun run build`: {e}. Install bun to build the frontend."
+            );
             ensure_dist_dir();
         }
     }

docs/content/docs/(configuration)/config.mdx

@@ -24,6 +24,7 @@ spacebot --config /path/to.toml  # CLI override
 anthropic_key = "env:ANTHROPIC_API_KEY"
 openai_key = "env:OPENAI_API_KEY"
 openrouter_key = "env:OPENROUTER_API_KEY"
+ollama_key = "env:OLLAMA_API_KEY"
 
 # --- Instance Defaults ---
 # All agents inherit these. Individual agents can override any field.
@@ -168,6 +169,7 @@ Model names include the provider as a prefix:
 | Anthropic | `anthropic/<model>` | `anthropic/claude-sonnet-4-20250514` |
 | OpenAI | `openai/<model>` | `openai/gpt-4o` |
 | OpenRouter | `openrouter/<provider>/<model>` | `openrouter/anthropic/claude-sonnet-4-20250514` |
+| Ollama Cloud | `ollama/<model>` | `ollama/gpt-oss:20b` |
 
 You can mix providers across process types. See [Routing](/docs/routing) for the full routing system.
 

docs/content/docs/(deployment)/roadmap.mdx

@@ -18,7 +18,7 @@ The full message-in → LLM → response-out pipeline is wired end-to-end across
 - **Config** — hierarchical TOML with `Config`, `AgentConfig`, `ResolvedAgentConfig`, `Binding`, `MessagingConfig`. File watcher with event filtering and content hash debounce for hot-reload.
 - **Multi-agent** — per-agent database isolation, `Agent` struct bundles all dependencies
 - **Database connections** — SQLite + LanceDB + redb per-agent, migrations for all tables
-- **LLM** — `SpacebotModel` implements Rig's `CompletionModel`, routes through `LlmManager` via HTTP with retries and fallback chains across 11 providers (Anthropic, OpenAI, OpenRouter, Z.ai, Groq, Together, Fireworks, DeepSeek, xAI, Mistral, OpenCode Zen)
+- **LLM** — `SpacebotModel` implements Rig's `CompletionModel`, routes through `LlmManager` via HTTP with retries and fallback chains across 12 providers (Anthropic, OpenAI, OpenRouter, Ollama Cloud, Z.ai, Groq, Together, Fireworks, DeepSeek, xAI, Mistral, OpenCode Zen)
 - **Model routing** — `RoutingConfig` with process-type defaults, task overrides, fallback chains
 - **Memory** — full stack: types, SQLite store (CRUD + graph), LanceDB (embeddings + vector + FTS), fastembed, hybrid search (RRF fusion). `memory_type` filter wired end-to-end through SearchConfig. `total_cmp` for safe sorting.
 - **Memory maintenance** — decay + prune implemented

interface/src/lib/providerIcons.tsx

@@ -60,6 +60,7 @@ export function ProviderIcon({ provider, className = "text-ink-faint", size = 24
 		anthropic: Anthropic,
 		openai: OpenAI,
 		openrouter: OpenRouter,
+		ollama: OpenRouter,
 		groq: Groq,
 		mistral: Mistral,
 		deepseek: DeepSeek,

interface/src/routes/Settings.tsx

@@ -95,6 +95,13 @@ const PROVIDERS = [
 		placeholder: "sk-...",
 		envVar: "OPENAI_API_KEY",
 	},
+	{
+		id: "ollama",
+		name: "Ollama Cloud",
+		description: "Hosted Ollama models via OpenAI-compatible API",
+		placeholder: "ollama_...",
+		envVar: "OLLAMA_API_KEY",
+	},
 	{
 		id: "zhipu",
 		name: "Z.ai (GLM)",

src/agent.rs

@@ -1,10 +1,10 @@
 //! Agent processes: channels, branches, workers, compactor, cortex.
 
-pub mod channel;
 pub mod branch;
-pub mod worker;
+pub mod channel;
 pub mod compactor;
 pub mod cortex;
 pub mod cortex_chat;
 pub mod ingestion;
 pub mod status;
+pub mod worker;

src/agent/branch.rs

@@ -2,10 +2,10 @@
 
 use crate::agent::compactor::estimate_history_tokens;
 use crate::error::Result;
-use crate::llm::routing::is_context_overflow_error;
-use crate::llm::SpacebotModel;
-use crate::{BranchId, ChannelId, ProcessId, ProcessType, AgentDeps, ProcessEvent};
 use crate::hooks::SpacebotHook;
+use crate::llm::SpacebotModel;
+use crate::llm::routing::is_context_overflow_error;
+use crate::{AgentDeps, BranchId, ChannelId, ProcessEvent, ProcessId, ProcessType};
 use rig::agent::AgentBuilder;
 use rig::completion::{CompletionModel, Prompt};
 use rig::tool::server::ToolServerHandle;
@@ -44,8 +44,14 @@ impl Branch {
     ) -> Self {
         let id = Uuid::new_v4();
         let process_id = ProcessId::Branch(id);
-        let hook = SpacebotHook::new(deps.agent_id.clone(), process_id, ProcessType::Branch, Some(channel_id.clone()), deps.event_tx.clone());
-
+        let hook = SpacebotHook::new(
+            deps.agent_id.clone(),
+            process_id,
+            ProcessType::Branch,
+            Some(channel_id.clone()),
+            deps.event_tx.clone(),
+        );
+
         Self {
             id,
             channel_id,
@@ -58,7 +64,7 @@ impl Branch {
             max_turns,
         }
     }
-    
+
     /// Run the branch's LLM agent loop and return a conclusion.
     ///
     /// Each branch has its own isolated ToolServer with `memory_save` and
@@ -70,7 +76,7 @@ impl Branch {
     /// be large, making them susceptible to overflow on the first LLM call.
     pub async fn run(mut self, prompt: impl Into<String>) -> Result<String> {
         let prompt = prompt.into();
-        
+
         tracing::info!(
             branch_id = %self.id,
             channel_id = %self.channel_id,
@@ -97,15 +103,17 @@ impl Branch {
         let mut overflow_retries = 0;
 
         let conclusion = loop {
-            match agent.prompt(&current_prompt)
+            match agent
+                .prompt(&current_prompt)
                 .with_history(&mut self.history)
                 .with_hook(self.hook.clone())
                 .await
             {
                 Ok(response) => break response,
                 Err(rig::completion::PromptError::MaxTurnsError { .. }) => {
-                    let partial = extract_last_assistant_text(&self.history)
-                        .unwrap_or_else(|| "Branch exhausted its turns without a final conclusion.".into());
+                    let partial = extract_last_assistant_text(&self.history).unwrap_or_else(|| {
+                        "Branch exhausted its turns without a final conclusion.".into()
+                    });
                     tracing::warn!(branch_id = %self.id, "branch hit max turns, returning partial result");
                     break partial;
                 }
@@ -133,7 +141,8 @@ impl Branch {
                         "branch context overflow, compacting and retrying"
                     );
                     self.force_compact_history();
-                    current_prompt = "Continue where you left off. Older context has been compacted.".into();
+                    current_prompt =
+                        "Continue where you left off. Older context has been compacted.".into();
                 }
                 Err(error) => {
                     tracing::error!(branch_id = %self.id, %error, "branch LLM call failed");
@@ -149,9 +158,9 @@ impl Branch {
             channel_id: self.channel_id.clone(),
             conclusion: conclusion.clone(),
         });
-        
+
         tracing::info!(branch_id = %self.id, "branch completed");
-        
+
         Ok(conclusion)
     }
 
@@ -192,7 +201,9 @@ impl Branch {
             return;
         }
 
-        let remove_count = ((total as f32 * fraction) as usize).max(1).min(total.saturating_sub(2));
+        let remove_count = ((total as f32 * fraction) as usize)
+            .max(1)
+            .min(total.saturating_sub(2));
         self.history.drain(..remove_count);
 
         let marker = format!(
@@ -207,7 +218,8 @@ impl Branch {
 fn extract_last_assistant_text(history: &[rig::message::Message]) -> Option<String> {
     for message in history.iter().rev() {
         if let rig::message::Message::Assistant { content, .. } = message {
-            let texts: Vec<String> = content.iter()
+            let texts: Vec<String> = content
+                .iter()
                 .filter_map(|c| {
                     if let rig::message::AssistantContent::Text(t) = c {
                         Some(t.text.clone())

src/agent/compactor.rs

@@ -25,11 +25,7 @@ pub struct Compactor {
 
 impl Compactor {
     /// Create a new compactor for a channel.
-    pub fn new(
-        channel_id: ChannelId,
-        deps: AgentDeps,
-        history: Arc<RwLock<Vec<Message>>>,
-    ) -> Self {
+    pub fn new(channel_id: ChannelId, deps: AgentDeps, history: Arc<RwLock<Vec<Message>>>) -> Self {
         Self {
             channel_id,
             deps,
@@ -117,12 +113,7 @@ impl Compactor {
             .expect("failed to render compactor prompt");
 
         tokio::spawn(async move {
-            let result = run_compaction(
-                &deps,
-                &compactor_prompt,
-                &history,
-                fraction,
-            ).await;
+            let result = run_compaction(&deps, &compactor_prompt, &history, fraction).await;
 
             match result {
                 Ok(turns_compacted) => {
@@ -191,7 +182,9 @@ async fn run_compaction(
     let (removed_messages, remove_count) = {
         let mut hist = history.write().await;
         let total = hist.len();
-        let remove_count = ((total as f32 * fraction) as usize).max(1).min(total.saturating_sub(2));
+        let remove_count = ((total as f32 * fraction) as usize)
+            .max(1)
+            .min(total.saturating_sub(2));
         if remove_count == 0 {
             return Ok(0);
         }
@@ -205,12 +198,14 @@ async fn run_compaction(
     // 3. Run the compaction LLM to produce summary + extracted memories
     let routing = deps.runtime_config.routing.load();
     let model_name = routing.resolve(ProcessType::Worker, None).to_string();
-    let model = SpacebotModel::make(&deps.llm_manager, &model_name)
-        .with_routing((**routing).clone());
+    let model =
+        SpacebotModel::make(&deps.llm_manager, &model_name).with_routing((**routing).clone());
 
     // Give the compaction worker memory_save so it can directly persist memories
     let tool_server: ToolServerHandle = ToolServer::new()
-        .tool(crate::tools::MemorySaveTool::new(deps.memory_search.clone()))
+        .tool(crate::tools::MemorySaveTool::new(
+            deps.memory_search.clone(),
+        ))
         .run();
 
     let agent = AgentBuilder::new(model)
@@ -220,7 +215,8 @@ async fn run_compaction(
         .build();
 
     let mut compaction_history = Vec::new();
-    let response = agent.prompt(&transcript)
+    let response = agent
+        .prompt(&transcript)
         .with_history(&mut compaction_history)
         .await;
 
@@ -294,9 +290,7 @@ fn estimate_assistant_content_chars(content: &AssistantContent) -> usize {
         AssistantContent::ToolCall(tc) => {
             tc.function.name.len() + tc.function.arguments.to_string().len()
         }
-        AssistantContent::Reasoning(r) => {
-            r.reasoning.iter().map(|s| s.len()).sum()
-        }
+        AssistantContent::Reasoning(r) => r.reasoning.iter().map(|s| s.len()).sum(),
         AssistantContent::Image(_) => 500,
     }
 }
@@ -339,8 +333,7 @@ fn render_messages_as_transcript(messages: &[Message]) -> String {
                         AssistantContent::ToolCall(tc) => {
                             output.push_str(&format!(
                                 "[Tool Call: {}({})]\n",
-                                tc.function.name,
-                                tc.function.arguments
+                                tc.function.name, tc.function.arguments
                             ));
                         }
                         _ => {}