@oh-my-pi/pi-agent-core 18.8.6 → 18.8.8

package/CHANGELOG.md

@@ -2,6 +2,19 @@
 
 ## [Unreleased]
 
+## [18.8.8] - 2026-10-10
+
+### Added
+
+- Added `Agent.setModelResolver()` to fit every model an agent adopts (via `setModel`, starting with the current one) before it is used ([#15048](https://github.com/can1357/oh-my-pi/pull/15048) by [@H4vC](https://github.com/H4vC))
+- Added `CompactionSettings.baseWindowTokens`: when set, `resolveThresholdTokens()` scales its percentage or reserve-based threshold from that base instead of the full context window ([#15065](https://github.com/can1357/oh-my-pi/pull/15065) by [@H4vC](https://github.com/H4vC))
+
+### Fixed
+
+- Fixed OpenAI V1 remote compaction re-attaching a prior Anthropic native payload after a successful compact ([#15041](https://github.com/can1357/oh-my-pi/pull/15041) by [@PaleRoses](https://github.com/PaleRoses))
+- Fixed `resolveThresholdTokens()` clamping a positive `thresholdTokens` to `baseWindowTokens`; a fixed threshold is checked against the real window, and the base only rescales the percentage and reserve policies ([#15074](https://github.com/can1357/oh-my-pi/pull/15074) by [@H4vC](https://github.com/H4vC))
+- Fixed `resolveThresholdTokens()` clamping a fixed `thresholdTokens` at or past the context window to one token below it, so compaction fired only after the next request overflowed; it now clamps to the window less the reserve ([#15146](https://github.com/can1357/oh-my-pi/pull/15146) by [@will-bogusz](https://github.com/will-bogusz))
+
 ## [18.8.6] - 2026-10-08
 
 ### Added

package/dist/types/agent.d.ts

@@ -469,6 +469,12 @@ export declare class Agent {
     emitExternalEvent(event: AgentEvent): void;
     setSystemPrompt(v: string[] | string): void;
     setModel(model: Model): void;
+    /**
+     * Route every model this agent adopts through `resolver` (e.g. to fit a
+     * shared catalog row to this agent's own settings), starting with the
+     * current one; `undefined` adopts models as given from now on.
+     */
+    setModelResolver(resolver: ((model: Model) => Model) | undefined): void;
     setThinkingLevel(l: Effort | undefined): void;
     setDisableReasoning(disabled: boolean): void;
     setSteeringMode(mode: "all" | "one-at-a-time"): void;

package/dist/types/compaction/compaction.d.ts

@@ -34,6 +34,14 @@ export interface CompactionSettings {
     strategy?: "context-full" | "handoff" | "shake" | "snapcompact" | "off";
     thresholdPercent?: number;
     thresholdTokens?: number;
+    /**
+     * Stands in for the context window when the percentage or reserve-based
+     * threshold is computed (when `> 0` and smaller than the window); a positive
+     * `thresholdTokens` still wins and is checked against the real window (at
+     * or past it, the window less its reserve applies). Set by per-model
+     * compaction limits; request and overflow budgets keep the real window.
+     */
+    baseWindowTokens?: number;
     midTurnEnabled?: boolean;
     /**
      * Tokens reserved below the context window for the next prompt + response.

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-agent-core",
-	"version": "18.8.6",
+	"version": "18.8.8",
 	"description": "General-purpose agent with transport abstraction, state management, and attachment support",
 	"homepage": "https://omp.sh",
 	"author": {
@@ -38,16 +38,16 @@
 		"fmt": "oxfmt --no-error-on-unmatched-pattern 'src/**/*.{ts,tsx}' '{test,bench,examples,scripts}/**/*.ts' '*.ts'"
 	},
 	"dependencies": {
-		"@oh-my-pi/pi-ai": "18.8.6",
-		"@oh-my-pi/pi-catalog": "18.8.6",
-		"@oh-my-pi/pi-natives": "18.8.6",
-		"@oh-my-pi/pi-utils": "18.8.6",
-		"@oh-my-pi/pi-wire": "18.8.6",
-		"@oh-my-pi/snapcompact": "18.8.6",
+		"@oh-my-pi/pi-ai": "18.8.8",
+		"@oh-my-pi/pi-catalog": "18.8.8",
+		"@oh-my-pi/pi-natives": "18.8.8",
+		"@oh-my-pi/pi-utils": "18.8.8",
+		"@oh-my-pi/pi-wire": "18.8.8",
+		"@oh-my-pi/snapcompact": "18.8.8",
 		"@opentelemetry/api": "^1.9.1"
 	},
 	"devDependencies": {
-		"@oh-my-pi/omptype": "18.8.6",
+		"@oh-my-pi/omptype": "18.8.8",
 		"@opentelemetry/context-async-hooks": "^2.9.0",
 		"@opentelemetry/sdk-trace-base": "^2.9.0",
 		"@types/bun": "^1.3.14"

package/src/agent.ts

@@ -453,6 +453,7 @@ export class Agent {
 	#promptCacheKey?: string;
 	#metadata?: Record<string, unknown>;
 	#metadataResolver?: (provider: string) => Record<string, unknown> | undefined;
+	#modelResolver?: (model: Model) => Model;
 	#providerSessionState?: Map<string, ProviderSessionState>;
 	#thinkingBudgets?: ThinkingBudgets;
 	#temperature?: number;
@@ -1213,8 +1214,20 @@ export class Agent {
 	}
 
 	setModel(model: Model) {
-		this.#state.model = model;
-		this.#syncTokenizer(model);
+		// Sessions may start with no model selected (`initialState.model: undefined`).
+		const resolved = this.#modelResolver && model ? this.#modelResolver(model) : model;
+		this.#state.model = resolved;
+		this.#syncTokenizer(resolved);
+	}
+
+	/**
+	 * Route every model this agent adopts through `resolver` (e.g. to fit a
+	 * shared catalog row to this agent's own settings), starting with the
+	 * current one; `undefined` adopts models as given from now on.
+	 */
+	setModelResolver(resolver: ((model: Model) => Model) | undefined): void {
+		this.#modelResolver = resolver;
+		if (resolver && this.#state.model) this.setModel(this.#state.model);
 	}
 
 	setThinkingLevel(l: Effort | undefined) {

package/src/compaction/compaction.ts

@@ -163,6 +163,14 @@ export interface CompactionSettings {
 	strategy?: "context-full" | "handoff" | "shake" | "snapcompact" | "off";
 	thresholdPercent?: number;
 	thresholdTokens?: number;
+	/**
+	 * Stands in for the context window when the percentage or reserve-based
+	 * threshold is computed (when `> 0` and smaller than the window); a positive
+	 * `thresholdTokens` still wins and is checked against the real window (at
+	 * or past it, the window less its reserve applies). Set by per-model
+	 * compaction limits; request and overflow budgets keep the real window.
+	 */
+	baseWindowTokens?: number;
 	midTurnEnabled?: boolean;
 	/**
 	 * Tokens reserved below the context window for the next prompt + response.
@@ -358,27 +366,42 @@ export function compactionContextTokens(providerContextTokens: number, storedCon
 	return Math.max(Math.max(0, providerContextTokens), Math.max(0, storedConversationEstimate));
 }
 
+/**
+ * The prompt budget below `contextWindow`: the window less
+ * {@link resolveBudgetReserveTokens}, capped at `contextWindow - 1` so it never
+ * reaches the whole window even when the reserve resolves to 0.
+ */
+function promptBudgetTokens(contextWindow: number, settings: CompactionSettings): number {
+	return Math.max(0, Math.min(contextWindow - 1, contextWindow - resolveBudgetReserveTokens(contextWindow, settings)));
+}
+
 export function resolveThresholdTokens(contextWindow: number, settings: CompactionSettings): number {
-	// Fixed token limit takes priority over percentage
+	// Fixed token limit takes priority over percentage, and is checked against
+	// the real window: `baseWindowTokens` only rescales the policies below.
 	const thresholdTokens = settings.thresholdTokens;
 	if (typeof thresholdTokens === "number" && Number.isFinite(thresholdTokens) && thresholdTokens > 0) {
-		// Clamp to [1, contextWindow - 1] so there's always room
+		// A trigger at or past the window can never fire before the request
+		// overflows (a provider may cap the window below the configured point);
+		// fall back to the window's prompt budget, as the reserve policy does.
+		// A trigger below the window is exact, as the `f` model entries promise,
+		// even one inside the reserve: the user chose that headroom. The step at
+		// the window is inherent; a monotone fallback would sit at window - 1.
+		if (thresholdTokens >= contextWindow) return promptBudgetTokens(contextWindow, settings);
 		return Math.min(contextWindow - 1, Math.max(1, thresholdTokens));
 	}
+	const baseWindowTokens = settings.baseWindowTokens;
+	if (typeof baseWindowTokens === "number" && Number.isFinite(baseWindowTokens) && baseWindowTokens > 0) {
+		contextWindow = Math.min(contextWindow, baseWindowTokens);
+	}
 
 	// Percentage-based threshold. The default absolute reserve can exceed bundled
 	// small-context windows, or nearly consume a 16k-class window; in those
 	// known-impossible default configurations, fall back to the proportional
 	// reserve so threshold/recovery-band checks stay usable. Explicit valid
-	// configured reserves still define the usable prompt budget. Cap at
-	// contextWindow - 1 (matching the fixed-token clamp above) so the threshold
-	// never reaches the whole window even when the reserve resolves to 0.
+	// configured reserves still define the usable prompt budget.
 	const thresholdPercent = settings.thresholdPercent;
 	if (typeof thresholdPercent !== "number" || !Number.isFinite(thresholdPercent) || thresholdPercent <= 0) {
-		return Math.max(
-			0,
-			Math.min(contextWindow - 1, contextWindow - resolveBudgetReserveTokens(contextWindow, settings)),
-		);
+		return promptBudgetTokens(contextWindow, settings);
 	}
 	const clampedThresholdPercent = Math.min(99, Math.max(1, thresholdPercent));
 	return Math.floor(contextWindow * (clampedThresholdPercent / 100));
@@ -1845,7 +1868,7 @@ export async function compact(
 						),
 					{ signal },
 				);
-				preserveData = withOpenAiRemoteCompactionPreserveData(previousPreserveData, remote);
+				preserveData = withOpenAiRemoteCompactionPreserveData(preserveData, remote);
 				usedRemoteCompaction = true;
 			} catch (err) {
 				// A user/session abort is a cancellation, not a remote failure —