model/context: per-model ctx window overrides + expose window size in /api/state

hive-ag3nt/src/events.rs

@@ -303,8 +303,30 @@ pub fn default_model() -> &'static str {
 /// Overridable at runtime via `HIVE_CONTEXT_WINDOW_TOKENS` (useful for
 /// future models or when the operator knows the exact limit). The env
 /// var takes precedence over the model-name heuristic.
+///
+/// Resolution order (first match wins):
+/// 1. `HIVE_CONTEXT_WINDOW_TOKENS_<KEY>` — per-model override where KEY
+///    (case-insensitive) is a substring of the active model name.
+///    Set by `hyperhive.contextWindowTokens.<key>` in `agent.nix`.
+/// 2. `HIVE_CONTEXT_WINDOW_TOKENS` — global override (any model).
+/// 3. Auto-derive: haiku → 200 000, sonnet / opus → 1 000 000.
 #[must_use]
 pub fn context_window_tokens(model: &str) -> u64 {
+    let m = model.to_ascii_lowercase();
+    // Per-model overrides: HIVE_CONTEXT_WINDOW_TOKENS_<KEY_UPPER> where
+    // KEY (lowercased) must be a non-empty substring of the model name.
+    for (key, val) in std::env::vars() {
+        if let Some(suffix) = key.strip_prefix("HIVE_CONTEXT_WINDOW_TOKENS_") {
+            if !suffix.is_empty() && m.contains(&suffix.to_ascii_lowercase()) {
+                if let Ok(v) = val.trim().parse::<u64>() {
+                    if v > 0 {
+                        return v;
+                    }
+                }
+            }
+        }
+    }
+    // Global override.
     if let Ok(s) = std::env::var("HIVE_CONTEXT_WINDOW_TOKENS") {
         if let Ok(v) = s.trim().parse::<u64>() {
             if v > 0 {
@@ -312,7 +334,7 @@ pub fn context_window_tokens(model: &str) -> u64 {
             }
         }
     }
-    let m = model.to_ascii_lowercase();
+    // Auto-derive from model family.
     if m.contains("sonnet") || m.contains("opus") {
         1_000_000
     } else {

hive-ag3nt/src/web_ui.rs

@@ -357,6 +357,12 @@ struct StateSnapshot {
     /// the operator can see what they just switched to (and what's
     /// in flight). Mutable at runtime via `POST /api/model`.
     model: String,
+    /// Effective context-window token budget for the current model.
+    /// Derived from `events::context_window_tokens(&model)` — respects
+    /// per-model and global `HIVE_CONTEXT_WINDOW_TOKENS_*` overrides then
+    /// falls back to model-family heuristic. Consumers (e.g. dashboard
+    /// badge) use this to render the ctx-usage percentage.
+    context_window_tokens: u64,
     /// Last-inference token usage from the most recent completed
     /// turn — represents the current context-window size at turn-end.
     /// `null` until the first turn finishes.
@@ -451,6 +457,7 @@ async fn api_state(State(state): State<AppState>) -> axum::Json<StateSnapshot> {
     let inbox = recent_inbox(&state.socket, state.flavor()).await;
     let (turn_state, turn_state_since) = state.bus.state_snapshot();
     let model = state.bus.model();
+    let context_window_tokens = crate::events::context_window_tokens(&model);
     let ctx_usage = state.bus.last_ctx_usage();
     let cost_usage = state.bus.last_cost_usage();
     axum::Json(StateSnapshot {
@@ -463,6 +470,7 @@ async fn api_state(State(state): State<AppState>) -> axum::Json<StateSnapshot> {
         turn_state,
         turn_state_since,
         model,
+        context_window_tokens,
         ctx_usage,
         cost_usage,
         gui_enabled: state.gui_vnc_port.is_some(),