pi-freeflow 1.33.2 → 1.34.0

package/CHANGELOG.md

@@ -1,5 +1,47 @@
 # Changelog
 
+## 1.34.0
+
+### Minor Changes
+
+- aa6f259: Three new free models, picked the same way as the rest (`/model` → `freeflow` → pick):
+
+  - **Step 5 Preview** on OpenCode Zen (`step-5-preview-free`, short name `step-5-preview`) — 1M context, answers with vision.
+  - **Step 5 Preview** on KiloCode (`stepfun/step-5-preview-free`, short name `step-5-preview:kilo`) — 1M context, answers with vision.
+  - **Glyph Cluster** on KiloCode (`stealth/glyph-cluster`, short name `glyph-cluster`) — 256K context, text answers.
+
+### Patch Changes
+
+- 50981fb: Align nine model entries with the specs their upstream catalogs publish. Models now advertise video input where the vendor lists it, two entries pick up a larger context window, and Step 5 Preview on KiloCode reports the full 64K output allowance.
+- f637c58: Recover tool calls that arrive with the tail of the previous call stuck to the
+  front. Some models open a new call by echoing the end of the one they have just
+  seen, so a single call is delivered as `<tail of previous>{"…this call…"}` and
+  the host refuses it as invalid JSON. When such a payload does not parse, the
+  proxy now locates the complete call behind the stray text and hands that back.
+
+  Only payloads that already fail to parse are rescanned, and only a position that
+  yields a complete object is accepted — a well-formed call is never rewritten,
+  and a truncated one is still refused rather than completed.
+
+  Also adds a whole-inventory fidelity test covering every tool either host can
+  declare, across all three API shapes.
+
+- aa6f259: Remove three free models that stopped serving upstream, so they no longer appear in the model list:
+
+  - **Fledge Alpha** on OpenCode (`fledge-alpha-free`) — withdrawn upstream and no longer listed.
+  - **Ling 3.0 Flash Sante** on KiloCode (`inclusionai/ling-3.0-flash-sante:free`) — the free variant is gone; only the paid one remains.
+  - **Step 3.7 Flash** on KiloCode (`stepfun/step-3.7-flash:free`) — the free variant is gone; only the paid one remains.
+
+  All three are excluded permanently, so a stale on-disk catalog or a later catalog refresh cannot bring them back.
+
+- 50981fb: Correct the OpenCode Zen Step 5 Preview entry: it is a reasoning model with 1M context, 64K max output, text/image/video input, and only `low`/`medium`/`high` thinking levels (`minimal` is explicitly disabled). Previously the entry mirrored the Kilo sibling and the fallback heuristic, so the picker showed no thinking support.
+
+## 1.33.3
+
+### Patch Changes
+
+- Fix tool calls being rejected with an invalid-arguments error when a model makes several calls in one turn. Some models label every parallel call `0`, so the pieces of different calls were joined together and arrived as one unreadable blob — a planning tool could then reject every update, even a minimal one, and leave its task list stuck mid-turn. Calls are now kept apart by their own identifier, and a call whose text arrives in several pieces still reassembles whole.
+
 ## 1.33.2
 
 ### Patch Changes

package/README.md

@@ -205,7 +205,7 @@ Good defaults for long coding sessions and agentic work.
 | `big-pickle` | Big Pickle | **200K** (200.000) | **32K** (32.000) | `high / max` | ❌ |
 | `space-bunny-free` | Stealth preview (lab undisclosed) | **1M** (1.048.576) | **512K** (524.288) | `low … max` | ✅ |
 | `longcat-2.5-preview-free` | Meituan LongCat | **1M** (1.000.000) | **131K** (131.072) | `minimal … max` | ✅ |
-| `fledge-alpha-free` | Stealth preview (lab undisclosed) | **1M** (1.048.576) | **131K** (131.072) | `low / high / max` | ✅ |
+| `step-5-preview-free` | StepFun | **1M** (1.000.000) | **64K** (65.536) | `low … high` | ✅ |
 | `ling-3.1-flash-free` | Inclusion AI | **262K** (262.144) | **32K** (32.768) | `minimal … max` | ❌ |
 
 #### KiloCode Gateway (16 models), OpenRouter compatible
@@ -225,8 +225,8 @@ Keyless access. Short aliases work for every row (the full ID is in parentheses)
 | `lfm-2.5` (`liquid/lfm-2.5-2.6b:free`) | Liquid AI | **65K** (65.536) | **32K** (32.768) | `minimal…xhigh`\* | ❌ |
 | `kilo-auto` (`kilo-auto/free`) | Kilo Gateway Auto | **256K** (256.000) | **10K** (10.000) | `minimal…xhigh`\* | ❌ |
 | `openrouter` (`openrouter/free`) | OpenRouter Free | **200K** (200.000) | **65K** (65.536) | `minimal…xhigh`\* | ✅ |
-| `ling-3.0-flash-sante` (`inclusionai/ling-3.0-flash-sante:free`) | Inclusion AI | **262K** (262.144) | **32K** (32.768) | `minimal…xhigh`\* | ❌ |
-| `step-3.7-flash` (`stepfun/...:free`) | StepFun | **262K** (262.144) | **262K** (262.144) | `minimal…xhigh`\* | ✅ |
+| `glyph-cluster` (`stealth/glyph-cluster`) | Stealth preview (lab undisclosed) | **256K** (256.000) | **256K** (256.000) | `minimal…xhigh`\* | ❌ |
+| `step-5-preview:kilo` (`stepfun/...:free`) | StepFun | **1M** (1.000.000) | **64K** (64.000) | `minimal…xhigh`\* | ✅ |
 | `inkling-small` (`thinkingmachines/inkling-small:free`) | Thinking Machines | **1M** (1.048.576) | **262K** (262.144) | `minimal…xhigh`\* | ✅ |
 | `ling-3.1-flash-kilo` (`inclusionai/ling-3.1-flash`) | Inclusion AI | **262K** (262.144) | **32K** (32.768) | `minimal…xhigh`\* | ❌ |
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "pi-freeflow",
   "type": "module",
-  "version": "1.33.2",
+  "version": "1.34.0",
   "description": "Thin provider for OMP/Pi — model list + dumb relay proxy + log; host pi-ai owns thinking/normalization",
   "main": "extensions/index.ts",
   "types": "src/index.ts",

package/src/catalog.ts

@@ -30,7 +30,8 @@ import type {
 
 /**
  * Pruned model IDs that must never re-enter the catalog via disk cache or upstream merge.
- * - jev-1.13-free: non-chat decision model, chat-incompatible (live chat 500) — never registered.
+ * - jev-1.13-free: non-chat decision model, chat-incompatible — never registered.
+ *   Re-confirmed 2026-10-09: 400 through the plugin while every catalog model 200s.
  * - deepseek-v4-flash-free: listed but unserved upstream (live chat 400 "Model is unavailable").
  * - nex-agi/nex-n2.5-pro:free + nex-agi/nex-n2.5-mini:free: gone 2026-09-28 — zero nex-agi
  *   IDs on the live Kilo list, keyless chat 404 "does not exist".
@@ -43,7 +44,18 @@ import type {
  * - inclusionai/ling-3.0-flash-fin:free: gone from Kilo (2026-10-01) — keyless chat 404
  *   "The requested model ... does not exist".
  * - mimo-v2.5-free: decommissioned upstream (2026-10-07) — HTTP 401 "Model mimo-v2.5-free is not supported".
- * - exo-free: unserved upstream (2026-10-07) — HTTP 503 "Error from provider (Console): Upstream request failed: Endpoint is unavailable.".
+ * - exo-free: deprecated upstream (2026-10-09) — HTTP 410 "Model exo-free has
+ *   been deprecated." (was 503 "Endpoint is unavailable" on 2026-10-07).
+ * - fledge-alpha-free: withdrawn upstream (2026-10-09) — absent from the live
+ *   Zen list, chat 401 "Model fledge-alpha-free is not supported".
+ * - inclusionai/ling-3.0-flash-sante:free: demoted to paid on Kilo (2026-10-09) —
+ *   :free ID gone from the live list, keyless chat 404 "does not exist";
+ *   only the paid counterpart remains.
+ * - stepfun/step-3.7-flash:free: demoted to paid on Kilo (2026-10-09) — same
+ *   shape: :free ID gone, keyless 404; paid ID remains.
+ * - meituan/longcat-2.0-free: demoted to paid on Kilo (2026-10-09) —
+ *   `meituan/longcat-2.0` remains live with isFree:false; keyless chat
+ *   answers 401 PAID_MODEL_AUTH_REQUIRED.
  * - stealth/space-bunny-alpha: gone from Kilo (2026-10-07) — keyless chat 404 "The requested model ... does not exist".
  * - qwen/qwen3.8-27b:free: gone from Kilo (2026-10-07) — keyless chat 404 "The requested model is currently unavailable.".
  * - nvidia/nemotron-3.5-content-safety:free: tool-less guardrail model (2026-10-08),
@@ -78,6 +90,9 @@ export const DEAD_MODEL_IDS = new Set<string>([
  "stealth/space-bunny-alpha",
  "qwen/qwen3.8-27b:free",
  "nvidia/nemotron-3.5-content-safety:free",
+ "fledge-alpha-free",
+ "inclusionai/ling-3.0-flash-sante:free",
+ "stepfun/step-3.7-flash:free",
 ]);
 /**
  * Free-tier allowlist for anything entering the picker via network or stale disk.

package/src/models.ts

@@ -1,7 +1,7 @@
 /**
  * Static model definitions and upstream routing catalogs for pi-freeflow
  *
- * Defines the 30 verified free models (live-verified 2026-10-08):
+ * Defines the 30 verified free models (live-verified 2026-10-09):
  * - 10 OpenCode Zen models (2 Responses API + 8 Chat Completions)
  * - 15 KiloCode Keyless Gateway models (OpenRouter format)
  * - 5 Cline direct-only models (per-user pool)
@@ -21,7 +21,7 @@ export const OPENCODE_MODELS: ModelDef[] = [
   contextWindow: 1_048_576,
   maxTokens: 131_072,
   api: "openai-responses",
-  input: ["text", "image"],
+ input: ["text", "image", "video"],
   thinkingLevelMap: {
    off: null,
    minimal: "minimal",
@@ -39,7 +39,7 @@ export const OPENCODE_MODELS: ModelDef[] = [
   contextWindow: 1_048_576,
   maxTokens: 131_072,
   api: "openai-responses",
-  input: ["text", "image"],
+ input: ["text", "image", "video"],
   thinkingLevelMap: {
    off: null,
    minimal: "minimal",
@@ -56,7 +56,7 @@ export const OPENCODE_MODELS: ModelDef[] = [
   reasoning: true,
   contextWindow: 1_048_576,
   maxTokens: 131_072,
-  input: ["text", "image"],
+ input: ["text", "image", "video"],
   thinkingLevelMap: {
    off: null,
    minimal: "low",
@@ -124,7 +124,7 @@ export const OPENCODE_MODELS: ModelDef[] = [
   reasoning: true,
   contextWindow: 1_048_576,
   maxTokens: 524_288,
-  input: ["text", "image"],
+ input: ["text", "image", "video"],
   thinkingLevelMap: {
    off: null,
    minimal: null,
@@ -153,21 +153,30 @@ export const OPENCODE_MODELS: ModelDef[] = [
   },
  },
  {
-  id: "fledge-alpha-free",
-  name: "Fledge Alpha Free [OpenCode]",
-  reasoning: true,
-  contextWindow: 1_048_576,
-  maxTokens: 131_072,
-  input: ["text", "image"],
-  thinkingLevelMap: {
-   off: null,
-   minimal: null,
-   low: "low",
-   medium: null,
-   high: "high",
-   xhigh: null,
-   max: "max",
-  },
+ // Added 2026-10-09: replaces fledge-alpha-free (withdrawn upstream —
+ // absent from the live list, chat 401 "not supported"). Live on the Zen
+ // list; gated-lane 403 matches the working models from a refused egress,
+ // so it rides the same fingerprint path. Specs per models.dev
+ // (opencode/step-5-preview-free) + StepFun docs, NOT the enrichModelDef
+ // substring heuristic (which yields reasoning:false / 262K / text-only):
+ // 1M ctx, 64K max output, text+image+video, efforts low/medium/high only.
+ // minimal is explicitly null (omitting the key leaves it selectable in
+ // the host); xhigh/max are unsupported upstream.
+ id: "step-5-preview-free",
+ name: "Step 5 Preview [OpenCode]",
+ reasoning: true,
+ contextWindow: 1_000_000,
+ maxTokens: 65_536,
+ input: ["text", "image", "video"],
+ thinkingLevelMap: {
+  off: null,
+  minimal: null,
+  low: "low",
+  medium: "medium",
+  high: "high",
+  xhigh: null,
+  max: null,
+ },
  },
  {
   id: "ling-3.1-flash-free",
@@ -191,9 +200,10 @@ export const OPENCODE_MODELS: ModelDef[] = [
 /**
  * Shared effort map for Kilo reasoning models — verified live 2026-08-29:
  * gateway accepts flat reasoning_effort minimal..xhigh for every reasoning
- * model; stepfun/step-3.7-flash measured monotonic 77→313 thinking chars
- * across minimal→xhigh. Declaring the map locks the picker (instead of
- * host guessing) and matches the OpenCode-model pattern.
+ * model; a StepFun reasoning model (step-3.7-flash, since retired from the
+ * free list) measured monotonic 77→313 thinking chars across minimal→xhigh.
+ * Declaring the map locks the picker (instead of host guessing) and matches
+ * the OpenCode-model pattern.
  */
 const KILO_REASONING_MAP: ThinkingLevelMap = {
  off: null,
@@ -225,7 +235,7 @@ export const KILO_MODELS: ModelDef[] = [
   reasoning: true,
   contextWindow: 256_000,
   maxTokens: 131_072,
-  input: ["text", "image"],
+ input: ["text", "image", "video"],
   thinkingFormat: "openrouter",
   thinkingLevelMap: KILO_REASONING_MAP,
  },
@@ -320,14 +330,15 @@ export const KILO_MODELS: ModelDef[] = [
   thinkingLevelMap: KILO_REASONING_MAP,
  },
  {
-  // Resurrected 2026-09-28: back on the live Kilo free list
-  // (isFree:true, 0/0 pricing, no expiry) + keyless chat 200.
-  id: "stepfun/step-3.7-flash:free",
-  name: "Step 3.7 Flash [Kilo]",
+  // Added 2026-10-09: replaces step-3.7-flash (demoted to paid upstream —
+  // :free ID 404s keyless). Live on Kilo free list (isFree:true, 0/0
+  // pricing, 1M ctx) + keyless chat served (429 provider concurrency).
+  id: "stepfun/step-5-preview-free",
+  name: "Step 5 Preview [Kilo]",
   reasoning: true,
-  contextWindow: 262_144,
-  maxTokens: 262_144,
-  input: ["text", "image"],
+  contextWindow: 1_000_000,
+ maxTokens: 65_536,
+ input: ["text", "image", "video"],
   thinkingFormat: "openrouter",
   thinkingLevelMap: KILO_REASONING_MAP,
  },
@@ -342,11 +353,13 @@ export const KILO_MODELS: ModelDef[] = [
   thinkingLevelMap: KILO_REASONING_MAP,
  },
  {
-  id: "inclusionai/ling-3.0-flash-sante:free",
-  name: "Ling 3.0 Flash Sante [Kilo]",
+  // Added 2026-10-09: live on Kilo free list (isFree:true, 0/0 pricing,
+  // 256K ctx) + keyless chat 200.
+  id: "stealth/glyph-cluster",
+  name: "Glyph Cluster [Kilo]",
   reasoning: true,
-  contextWindow: 262_144,
-  maxTokens: 32_768,
+  contextWindow: 256_000,
+  maxTokens: 256_000,
   input: ["text"],
   thinkingFormat: "openrouter",
   thinkingLevelMap: KILO_REASONING_MAP,
@@ -418,7 +431,7 @@ export const CLINE_MODELS: ModelDef[] = [
   contextWindow: 1_048_576,
   maxTokens: 131_072,
   api: "openai-responses",
-  input: ["text", "image"],
+ input: ["text", "image", "video"],
   thinkingLevelMap: {
    off: null,
    minimal: "minimal",
@@ -433,9 +446,9 @@ export const CLINE_MODELS: ModelDef[] = [
   id: "z-ai/glm-5.3-flash",
   name: "GLM 5.3 Flash [Cline]",
   reasoning: true,
-  contextWindow: 1_000_000,
+ contextWindow: 1_048_576,
   maxTokens: 131_072,
-  input: ["text", "image"],
+ input: ["text", "image", "video"],
   thinkingLevelMap: CLINE_FLASH_REASONING_MAP,
  },
  {
@@ -448,7 +461,7 @@ export const CLINE_MODELS: ModelDef[] = [
   reasoning: true,
   contextWindow: 1_048_576,
   maxTokens: 131_072,
-  input: ["text", "image"],
+ input: ["text", "image", "video"],
   thinkingLevelMap: CLINE_FLASH_REASONING_MAP,
  },
  {
@@ -477,14 +490,14 @@ export const MODEL_ALIASES: Record<string, string> = {
  "nemotron-3-super": "nvidia/nemotron-3-super-120b-a12b:free",
  "north-mini-code": "cohere/north-mini-code:free",
  "lfm-2.5": "liquid/lfm-2.5-2.6b:free",
- "ling-3.0-flash-sante": "inclusionai/ling-3.0-flash-sante:free",
- "step-3.7-flash": "stepfun/step-3.7-flash:free",
+ "step-5-preview:kilo": "stepfun/step-5-preview-free",
+ "glyph-cluster": "stealth/glyph-cluster",
  "inkling-small": "thinkingmachines/inkling-small:free",
  "mimo-v2.6-flash": "mimo-v2.6-flash-free",
  "space-bunny": "space-bunny-free",
  "longcat-2.5-preview": "longcat-2.5-preview-free",
  "longcat": "longcat-2.5-preview-free",
- "fledge-alpha": "fledge-alpha-free",
+ "step-5-preview": "step-5-preview-free",
  "ling-3.1-flash": "ling-3.1-flash-free",
  "ling-3.1-flash:kilo": "inclusionai/ling-3.1-flash",
  // provider-prefixed short aliases (slash-normalized)

package/src/opencode-fingerprint.ts

@@ -564,10 +564,16 @@ export function sseToChatCompletionJson(
  let reasoningContent = "";
  let role = "assistant";
  let usage: unknown = undefined;
+ // Keyed by tool-call id whenever the provider sends one. Index is NOT
+ // trustworthy: free models routinely label every parallel call `index: 0`,
+ // and merging their fragments produces one unparseable argument blob
+ // (observed live: `{"i":"a","op":"done"}{"i":"b","op":"done"}`). Index is
+ // only the fallback for providers that omit an id entirely.
  const toolCallsMap = new Map<
-  number,
+  string,
   { id: string; type: string; function: { name: string; arguments: string } }
  >();
+ const keyByIndex = new Map<number, string>();
 
  for (const ev of events) {
   if (ev.data === "[DONE]") continue;
@@ -591,8 +597,15 @@ export function sseToChatCompletionJson(
       }
       if (Array.isArray(delta.tool_calls)) {
        for (const tc of delta.tool_calls) {
-        const idx = tc.index ?? 0;
-        const existing = toolCallsMap.get(idx) ?? {
+        // The first delta of a call carries its id; later deltas carry only
+        // `arguments`. A later id-less delta therefore belongs to whatever
+        // call this index already opened.
+        const idx = typeof tc.index === "number" ? tc.index : 0;
+        const hasId = typeof tc.id === "string" && tc.id !== "";
+        let key = hasId ? `id:${tc.id}` : keyByIndex.get(idx);
+        if (key === undefined) key = `idx:${idx}`;
+        if (hasId) keyByIndex.set(idx, key);
+        const existing = toolCallsMap.get(key) ?? {
          id: tc.id || "",
          type: tc.type || "function",
          function: { name: "", arguments: "" },
@@ -601,7 +614,7 @@ export function sseToChatCompletionJson(
         if (tc.type) existing.type = tc.type;
         if (tc.function?.name) existing.function.name += tc.function.name;
         if (tc.function?.arguments) existing.function.arguments += tc.function.arguments;
-        toolCallsMap.set(idx, existing);
+        toolCallsMap.set(key, existing);
        }
       }
      }
@@ -617,9 +630,9 @@ export function sseToChatCompletionJson(
   }
  }
 
- const toolCalls = Array.from(toolCallsMap.entries())
-  .sort(([a], [b]) => a - b)
-  .map(([, tc]) => tc);
+ // Map insertion order is emission order, which is what the host expects. The
+ // previous numeric sort assumed the keys were indices; they are now ids.
+ const toolCalls = Array.from(toolCallsMap.values());
 
  const message: Record<string, unknown> = {
   role,

package/src/tool-args.ts

@@ -342,6 +342,46 @@ export function buildArgRepairSchemas(tools: unknown[]): Map<string, Record<stri
   return schemas;
 }
 
+
+/**
+ * Parse a tool-call argument string, salvaging one that arrives with a leading
+ * fragment of the *previous* call spliced onto its front.
+ *
+ * Observed live on space-bunny-free: a model that emitted a call, saw the
+ * result, then opened its next call by echoing the tail of the old one. The
+ * stream carried a single tool call, so keying fragments by call id cannot
+ * separate them, and the host rejected the payload:
+ *
+ *   `,"timeout":400}{"i":"…","command":"…"}`
+ *   `]}]}{"i":"Planning research phases","op":"init",…}`
+ *
+ * Both are one complete JSON object with junk in front of it. Only strings that
+ * do not parse as-is are considered, and only a position that yields a complete
+ * object is accepted, so a well-formed call never reaches this path and a
+ * truncated one is still refused rather than completed.
+ *
+ * Returns undefined when nothing salvageable is present.
+ */
+function parseArguments(argsJson: string): { value: Record<string, unknown>; salvaged: boolean } | undefined {
+  try {
+    const value = JSON.parse(argsJson);
+    if (value === null || typeof value !== "object" || Array.isArray(value)) return undefined;
+    return { value: value as Record<string, unknown>, salvaged: false };
+  } catch {
+    // fall through to the salvage scan
+  }
+  for (let i = argsJson.indexOf("{"); i !== -1; i = argsJson.indexOf("{", i + 1)) {
+    try {
+      const value = JSON.parse(argsJson.slice(i));
+      if (value !== null && typeof value === "object" && !Array.isArray(value)) {
+        return { value: value as Record<string, unknown>, salvaged: true };
+      }
+    } catch {
+      // not a start position; try the next brace
+    }
+  }
+  return undefined;
+}
 /**
  * Repair one tool call's arguments against its declared schema. `name` is the
  * downstream (restored) caller tool name. Returns the input string untouched
@@ -356,21 +396,20 @@ export function repairToolArguments(
   if (schemas === undefined || schemas.size === 0) return argsJson;
   const schema = schemas.get(name.trim().toLowerCase());
   if (schema === undefined) return argsJson;
-  let parsed: unknown;
-  try {
-   parsed = JSON.parse(argsJson);
-  } catch {
-   // Truncated or non-JSON arguments: paper over nothing.
-   return argsJson;
-  }
-  if (parsed === null || typeof parsed !== "object" || Array.isArray(parsed)) return argsJson;
-  const rec = parsed as Record<string, unknown>;
+  const parsed = parseArguments(argsJson);
+  // Nothing parses: truncated or non-JSON arguments. Paper over nothing.
+  if (parsed === undefined) return argsJson;
+  const rec = parsed.value;
   const repaired = repairValue(rec, schema, 0);
   const value = repaired.value === null || typeof repaired.value !== "object" || Array.isArray(repaired.value)
    ? rec
    : (repaired.value as Record<string, unknown>);
   const promoted = promoteStringArray(value, schema, 0);
   const result = promoted === value ? value : promoted;
+  // A salvaged payload needs no further reshaping: what follows the junk is
+  // already the caller's own object, and re-serializing it would be a change
+  // the repair was never asked to make.
+  if (parsed.salvaged) return matchesSchema(rec, schema, 0) ? JSON.stringify(rec) : argsJson;
   if (!repaired.changed && promoted === value) return argsJson;
   if (!matchesSchema(result, schema, 0)) return argsJson;
   return JSON.stringify(result);

package/src/types.ts

@@ -16,7 +16,7 @@ export type ThinkingLevel =
 
 export type ThinkingLevelMap = Partial<Record<ThinkingLevel, string | null>>;
 
-export type ModelInputType = "text" | "image";
+export type ModelInputType = "text" | "image" | "video";
 
 export interface ModelDef {
  id: string;