pi-fireworks-provider 1.5.0 → 1.5.2

package/README.md

@@ -15,7 +15,7 @@ _Kimi, MiniMax, GLM, DeepSeek, GPT-OSS — via Fireworks AI's Anthropic Messages
 
 ## Features
 
-- **39+ AI Models** including Kimi K2.5, MiniMax M2.5, GLM 4.5/4.7/5, DeepSeek V3.1/V3.2, DeepSeek V4 Flash, and GPT-OSS
+- **41+ AI Models** including Kimi K2.5, MiniMax M2.5, GLM 4.5/4.7/5, DeepSeek V3.1/V3.2, DeepSeek V4 Flash, and GPT-OSS
 - **Dual API support** via Fireworks AI's Anthropic Messages and OpenAI-compatible completions endpoints (per-model routing, matching pi core's Fireworks provider)
 - **Service tiers** — toggle Fireworks `priority` vs `standard` per request on supported models (with priority pricing reflected in cost tracking), via a keybinding, `/fireworks-tier`, and a footer status area
 - **Preserved thinking** — toggle Fireworks' `reasoning_history: "preserved"` so prior assistant reasoning is retained across turns (better multi-turn recall; uses more tokens), via the `/fireworks-settings` panel, with a model-select notification. Matches neuralwatt/makora's settings-only UX, adapted to Fireworks' single global `reasoning_history` knob
@@ -90,6 +90,7 @@ pi
 | GLM 5.2 Fast | Text | 1.0M | 131K | $2.10 | $6.60 |
 | GPT OSS 120B | Text | 131K | 33K | $0.15 | $0.60 |
 | GPT OSS 20B | Text | 131K | 33K | $0.07 | $0.30 |
+| Inkling | Text + Image | 1.0M | 0 | Free | Free |
 | Kimi K2 Instruct | Text | 131K | 16K | $1.00 | $3.00 |
 | Kimi K2 Thinking | Text | 262K | 256K | $0.60 | $2.50 |
 | Kimi K2.5 | Text + Image | 262K | 256K | $0.60 | $3.00 |
@@ -100,6 +101,7 @@ pi
 | Kimi K2.6 Turbo | Text + Image | 262K | 262K | $2.00 | $8.00 |
 | Kimi K2.7 Code | Text + Image | 262K | 262K | $0.95 | $4.00 |
 | Kimi K2.7 Code Fast | Text + Image | 262K | 262K | $1.90 | $8.00 |
+| Kimi K3 | Text + Image | 1.0M | 0 | Free | Free |
 | Llama 3.3 70B Instruct | Text | 131K | 0 | Free | Free |
 | MiniMax M2.7 (router) | Text | 204K | 0 | $0.30 | $1.20 |
 | MiniMax-M2.1 | Text | 197K | 200K | $0.30 | $1.20 |

package/deprecated-models.json

@@ -0,0 +1 @@
+{}

package/index.ts

@@ -26,10 +26,11 @@
  */
 
 import { getAgentDir, type ExtensionAPI, type ModelRegistry } from "@earendil-works/pi-coding-agent";
-import type { Input, matchesKey, Key, truncateToWidth, visibleWidth, wrapTextWithAnsi, fuzzyFilter, SettingsListTheme } from "@earendil-works/pi-tui";
+import type { Input, matchesKey, Key, KeyId, truncateToWidth, visibleWidth, wrapTextWithAnsi, fuzzyFilter, SettingsListTheme } from "@earendil-works/pi-tui";
 import modelsData from "./models.json" with { type: "json" };
 import customModelsData from "./custom-models.json" with { type: "json" };
 import patchData from "./patch.json" with { type: "json" };
+import deprecatedData from "./deprecated-models.json" with { type: "json" };
 import fs from "fs";
 import path from "path";
 
@@ -44,7 +45,7 @@ interface JsonModel {
   baseUrl?: string;
   reasoning: boolean;
   thinkingLevelMap?: Record<string, string | null>;
-  input: string[];
+  input: ("text" | "image")[];
   cost: {
     input: number;
     output: number;
@@ -64,7 +65,7 @@ interface PatchEntry {
   baseUrl?: string;
   reasoning?: boolean;
   thinkingLevelMap?: Record<string, string | null>;
-  input?: string[];
+  input?: ("text" | "image")[];
   cost?: {
     input?: number;
     output?: number;
@@ -118,7 +119,10 @@ function applyPatch(model: JsonModel, patch: PatchEntry): JsonModel {
 function buildModels(base: JsonModel[], custom: JsonModel[], patch: PatchData): JsonModel[] {
   const modelMap = new Map<string, JsonModel>();
 
-  for (const model of base) {
+  // Seed with the base list plus grace-period deprecated models so patch.json
+  // entries apply to deprecated models exactly as while the model was live
+  // (withDeprecated keeps live data on id conflicts).
+  for (const model of withDeprecated(base)) {
     modelMap.set(model.id, model);
   }
 
@@ -245,6 +249,35 @@ function mergeWithEmbedded(liveModels: JsonModel[], embeddedModels: JsonModel[])
   return result;
 }
 
+// Grace period for delisted models. When the provider API stops listing a
+// model, update-models.js moves its last-known definition into
+// deprecated-models.json (stamped with deprecatedAt) instead of dropping it.
+// For 14 days the model keeps working here so in-flight sessions and saved
+// model settings do not break; afterwards it is evicted permanently.
+const DEPRECATED_MODEL_TTL_MS = 14 * 24 * 60 * 60 * 1000;
+
+// Grace-period deprecated models with deprecation metadata stripped.
+function activeDeprecatedModels(): JsonModel[] {
+  const now = Date.now();
+  const result: JsonModel[] = [];
+  for (const entry of Object.values(deprecatedData as Record<string, JsonModel & { deprecatedAt?: string }>)) {
+    if (!entry?.id) continue;
+    const removedAt = Date.parse(entry.deprecatedAt ?? "");
+    if (Number.isNaN(removedAt) || now - removedAt > DEPRECATED_MODEL_TTL_MS) continue;
+    const model = { ...entry } as JsonModel & { deprecatedAt?: string };
+    delete model.deprecatedAt;
+    result.push(model);
+  }
+  return result;
+}
+
+// Append grace-period deprecated models the list does not already have (live data wins).
+function withDeprecated(models: JsonModel[]): JsonModel[] {
+  const seen = new Set(models.map((m) => m.id));
+  const extras = activeDeprecatedModels().filter((m) => !seen.has(m.id));
+  return extras.length > 0 ? [...models, ...extras] : models;
+}
+
 function loadStaleModels(embeddedModels: JsonModel[]): JsonModel[] {
   const cached = loadCachedModels();
   if (!cached || cached.length === 0) return embeddedModels;
@@ -371,7 +404,7 @@ type PreserveMode = boolean; // true = inject reasoning_history:"preserved"; fal
 
 interface ServiceTierConfig {
   default: ServiceTier;
-  keybinding: string;
+  keybinding: KeyId;
   display: "statusbar" | "off";
 }
 
@@ -425,7 +458,7 @@ function isValidTier(v: unknown): v is ServiceTier {
   return v === "standard" || v === "priority";
 }
 
-function isValidKeybinding(v: unknown): v is string {
+function isValidKeybinding(v: unknown): v is KeyId {
   return typeof v === "string" && v.length > 0;
 }
 

package/models.json

@@ -191,6 +191,23 @@
     "contextWindow": 131072,
     "maxTokens": 0
   },
+  {
+    "id": "accounts/fireworks/models/inkling",
+    "name": "Inkling",
+    "reasoning": false,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 0,
+      "output": 0,
+      "cacheRead": 0,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 0
+  },
   {
     "id": "accounts/fireworks/models/kimi-k2-instruct",
     "name": "Kimi K2 Instruct",
@@ -274,6 +291,23 @@
     "contextWindow": 262144,
     "maxTokens": 0
   },
+  {
+    "id": "accounts/fireworks/models/kimi-k3",
+    "name": "Kimi K3",
+    "reasoning": false,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 0,
+      "output": 0,
+      "cacheRead": 0,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 0
+  },
   {
     "id": "accounts/fireworks/models/llama-v3p3-70b-instruct",
     "name": "Llama 3.3 70B Instruct",
@@ -343,8 +377,7 @@
     "name": "Minimax M3",
     "reasoning": false,
     "input": [
-      "text",
-      "image"
+      "text"
     ],
     "cost": {
       "input": 0,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-fireworks-provider",
-  "version": "1.5.0",
+  "version": "1.5.2",
   "description": "Fireworks AI provider extension for pi - Access Kimi, MiniMax, GLM, DeepSeek, and GPT-OSS models through the Fireworks AI API",
   "type": "module",
   "main": "index.ts",
@@ -22,6 +22,7 @@
   "files": [
     "index.ts",
     "models.json",
+    "deprecated-models.json",
     "custom-models.json",
     "patch.json",
     "README.md",