integrations: fix four things that make openzoo fail out of the box

integrations/AnythingLLM/README.md

@@ -8,10 +8,10 @@ configuration — no file to install, so this folder is documentation only.
 
 1. Start the openzoo proxy: `npx openzoo`.
 2. In AnythingLLM: **Settings → AI Providers → LLM** → choose **Generic OpenAI**.
-3. **Base URL**: `http://localhost:8402/v1` (hosted: `https://api.openzoo.fun/v1`)
+3. **Base URL**: `http://localhost:8402/v1` (hosted: `https://x402-tokens.fly.dev/v1`)
 4. **API Key**: `sk-openzoo` (any value for local; real key for hosted)
 5. **Chat Model Name**: a zoo model id, e.g. `nvidia/nemotron-3.5-lightning`
-   (`npx openzoo models` lists ids and pricing)
+   (`curl localhost:8402/v1/models` lists ids and pricing)
 6. Set token context window / max tokens as desired, save, and chat.
 
 ## Notes

integrations/Cline/README.md

@@ -10,10 +10,10 @@ install, so this folder is documentation only.
 2. In Cline, click the ⚙️ settings icon.
 3. **API Provider**: select `OpenAI Compatible`.
 4. **Base URL**: `http://localhost:8402/v1`
-   (hosted: `https://api.openzoo.fun/v1`) — do NOT paste `/chat/completions`.
+   (hosted: `https://x402-tokens.fly.dev/v1`) — do NOT paste `/chat/completions`.
 5. **API Key**: `sk-openzoo` (any value for local; real key for hosted).
 6. **Model ID**: a zoo model id, e.g. `nvidia/nemotron-3.5-lightning`
-   — must exactly match an id returned by `/v1/models` (`npx openzoo models`).
+   — must exactly match an id returned by `/v1/models` (`curl localhost:8402/v1/models`).
 7. Fill the **Model Configuration** block (context window, output limits,
    whether the model supports images/computer-use). People skip this; for
    custom endpoints it matters — set the context window generously, since the

integrations/Continue/README.md

@@ -13,7 +13,7 @@ OpenAI-compatible endpoint via its built-in `provider: openai` — no plugin req
 ## Notes
 
 - Continue does not fetch a live model list; each zoo model you want is one entry.
-  `npx openzoo models` prints available model ids and pricing.
+  `curl localhost:8402/v1/models` prints available model ids and pricing.
 - `apiKey` is ignored by the local proxy (x402 pays); required-but-arbitrary. For a
   hosted endpoint, use your real key.
 - Coding sessions can push large repo context; the zoo's leCore corpus handling is

integrations/Continue/config.openzoo.yaml

@@ -4,7 +4,7 @@
 # Continue's native `provider: openai` speaks to any OpenAI-compatible
 # apiBase - no plugin needed. Each model you want available in Continue
 # gets its own entry (Continue has no live model-list fetch), so list the
-# zoo models you actually use. `npx openzoo models` prints the full list.
+# zoo models you actually use. `curl localhost:8402/v1/models` prints the full list.
 
 name: openzoo
 version: 0.0.1
@@ -17,7 +17,7 @@ models:
   - name: zoo nvidia/nemotron-3.5-lightning
     provider: openai
     model: nvidia/nemotron-3.5-lightning
-    apiBase: http://localhost:8402/v1     # hosted: https://api.openzoo.fun/v1
+    apiBase: http://localhost:8402/v1     # hosted: https://x402-tokens.fly.dev/v1
     apiKey: sk-openzoo                    # any value for local proxy; real key for hosted
     roles:
       - chat

integrations/Cursor/README.md

@@ -11,12 +11,12 @@ so this folder is documentation only.
 3. Disable pre-enabled Cursor-hosted models to avoid routing conflicts.
 4. In the OpenAI section, enable **Override OpenAI Base URL** and enter:
    `http://localhost:8402/v1`
-   (hosted: `https://api.openzoo.fun/v1`) — include the `/v1` suffix; Cursor
+   (hosted: `https://x402-tokens.fly.dev/v1`) — include the `/v1` suffix; Cursor
    appends `/chat/completions` itself.
 5. **OpenAI API Key**: `sk-openzoo` (any value for local; real key for hosted).
    Despite the "OpenAI" label, this key is sent to the overridden endpoint.
 6. Click **+ Add Model** and type a zoo model id, e.g. `nvidia/nemotron-3.5-lightning`
-   (`npx openzoo models` lists ids). Verify/Save — Cursor sends a test request.
+   (`curl localhost:8402/v1/models` lists ids). Verify/Save — Cursor sends a test request.
 7. Pick the zoo model in the chat sidebar (`Cmd/Ctrl+L`).
 
 ## Known limitations (Cursor-side, not openzoo-side)
@@ -36,7 +36,7 @@ so this folder is documentation only.
 ## Full platform access: add the MCP server too
 
 The chat override above covers ordinary completions. openzoo's biggest capability —
-`zoo_ask`, which takes a corpus of up to ~1M tokens (a body models refuse directly)
+`zoo_ask`, which takes a corpus of up to ~9.8M tokens per call (a body models refuse directly)
 and answers via leCore memory spill — is an **MCP tool**, and Cursor supports MCP.
 Settings → MCP → *Add new global MCP server*, or drop into `~/.cursor/mcp.json`
 (per-project: `.cursor/mcp.json`):

integrations/GrokCLI/README.md

@@ -3,7 +3,7 @@
 Grok CLI (superagent-ai/grok-cli — the open-source terminal coding agent built
 for the Grok API) speaks the OpenAI chat-completions format and accepts a custom
 base URL, so it can route through openzoo. This covers "Grok-ecosystem" users:
-they keep their harness and gain the zoo's ~435 models and leCore large-context
+they keep their harness and gain the zoo's 480+ models and leCore large-context
 handling.
 
 ## Setup
@@ -23,7 +23,7 @@ grok --model nvidia/nemotron-3.5-lightning
 
 Persistent — add the provider entry from `models.openzoo.json` in this folder to
 `~/.grok/models.json` (append to the existing array; add more zoo ids to
-`models` as desired — `npx openzoo models` lists them). Per-project model
+`models` as desired — `curl localhost:8402/v1/models` lists them). Per-project model
 pinning lives in `.grok/settings.json`.
 
 ## Notes

integrations/Hermes/README.md

@@ -11,7 +11,7 @@ Easiest: run `npx openzoo` (the paying proxy), then `hermes model`, choose
 
 - Base URL: `http://localhost:8402/v1` (stop at `/v1` — Hermes appends the route)
 - API key: `sk-openzoo` (any value for the local proxy; real key for hosted)
-- Model: any zoo id (`npx openzoo models` lists them)
+- Model: any zoo id (`curl localhost:8402/v1/models` lists them)
 
 Hermes saves the selection to `~/.hermes/config.yaml`. `config.openzoo.yaml` in
 this folder is the equivalent manual merge if you prefer editing config directly.

integrations/Hermes/config.openzoo.yaml

@@ -8,7 +8,7 @@
 #      -> base URL:  http://localhost:8402/v1   (ends in /v1; Hermes appends
 #                                                /chat/completions itself)
 #      -> API key:   sk-openzoo                 (any value for the local proxy)
-#      -> model id:  nvidia/nemotron-3.5-lightning          (or any zoo id; `npx openzoo models`)
+#      -> model id:  nvidia/nemotron-3.5-lightning          (or any zoo id; `curl localhost:8402/v1/models`)
 #
 #   B) Manual: merge the block below into ~/.hermes/config.yaml.
 #
@@ -19,5 +19,5 @@
 model:
   provider: custom
   name: nvidia/nemotron-3.5-lightning
-  base_url: http://localhost:8402/v1     # hosted: https://api.openzoo.fun/v1
+  base_url: http://localhost:8402/v1     # hosted: https://x402-tokens.fly.dev/v1
   api_key: sk-openzoo                    # any value for local proxy; real key for hosted

integrations/LibreChat/librechat.openzoo.yaml

@@ -15,7 +15,7 @@ endpoints:
   custom:
     - name: "openzoo"
       # Local proxy (start with: npx openzoo). For hosted, e.g.:
-      # baseURL: "https://api.openzoo.fun/v1"
+      # baseURL: "https://x402-tokens.fly.dev/v1"
       baseURL: "http://localhost:8402/v1"
       apiKey: "sk-openzoo"          # or "${OPENZOO_KEY}" for hosted
       models:

integrations/OpenWebUI/README.md

@@ -30,7 +30,7 @@ pays per call via x402; the key field is sent but ignored.
 
 Open the function's **Valves** and set:
 
-- `OPENZOO_BASE_URL` → the hosted URL (e.g. `https://api.openzoo.fun/v1`)
+- `OPENZOO_BASE_URL` → the hosted URL (e.g. `https://x402-tokens.fly.dev/v1`)
 - `OPENZOO_API_KEY` → your real key
 
 Nothing else changes; the same file serves both rails.

integrations/OpenWebUI/openzoo_pipe.py

@@ -5,7 +5,7 @@
 funding_url: https://openzoo.fun
 version: 0.1.0
 license: MIT
-description: Route chats through openzoo.fun - leCore-backed inference, ~435 models, effectively unlimited context, pay-per-call. Works with the local x402 proxy (npx openzoo) today and the hosted endpoint when available.
+description: Route chats through openzoo.fun - leCore-backed inference, 480+ models, effectively unlimited context, pay-per-call. Works with the local x402 proxy (npx openzoo) today and the hosted endpoint when available.
 requirements: requests
 """
 
@@ -31,8 +31,12 @@
 
 class Pipe:
     class Valves(BaseModel):
-        # Default is the local proxy started by `npx openzoo`. Point this at
-        # the hosted endpoint (e.g. https://api.openzoo.fun/v1) once it exists.
+        # Default is the local proxy started by `npx openzoo`. The hosted endpoint is
+        # https://x402-tokens.fly.dev/v1 (api.openzoo.fun/v1 404s -- it is the
+        # website, not the gateway).
+        # NOTE if you run open-webui in Docker (the documented install): set
+        # this to http://host.docker.internal:8402/v1 -- inside a container
+        # "localhost" is the container, and the proxy is on the HOST.
         OPENZOO_BASE_URL: str = Field(
             default="http://localhost:8402/v1",
             description="OpenAI-compatible base URL for openzoo (local proxy or hosted).",
@@ -95,17 +99,48 @@ def _strip_owui_prefix(self, model_id: str) -> str:
     })
 
     @staticmethod
-    def _format_receipt(usage: dict) -> str:
-        # openzoo returns billing info alongside usage; surface it so users
-        # see the per-call cost and the savings from the leCore spill.
+    def _format_receipt(data: dict) -> str:
+        # FIXED (was reading the wrong object): billing does NOT live in
+        # `usage`. Verified against the live gateway, a response carries a
+        # TOP-LEVEL "x402" block:
+        #   x402: {billedUsd, cogsUsd, directUsd, savesVsDirect,
+        #          subscription: {tier, cogsUsd, wouldHaveBilled, invoiced}}
+        # while `usage` holds only OpenAI/OpenRouter fields (prompt_tokens,
+        # completion_tokens, cost, cost_details, is_byok). The previous
+        # version looked for billedUsd/savesVsDirect inside `usage`, never
+        # found them, and silently rendered a token count and nothing else --
+        # i.e. the receipt, which is the whole reason this plugin exists, was
+        # dead code.
+        x = data.get("x402") or {}
+        usage = data.get("usage") or {}
+        sub = x.get("subscription") or {}
         parts = []
-        if "billedUsd" in usage:
-            parts.append(f"billed ${usage['billedUsd']}")
-        if "savesVsDirect" in usage:
-            parts.append(f"saved {usage['savesVsDirect']} vs direct")
-        if "prompt_tokens" in usage:
-            parts.append(f"{usage['prompt_tokens']} tokens read")
-        return ("\n\n---\n*openzoo: " + ", ".join(parts) + "*") if parts else ""
+
+        billed = x.get("billedUsd")
+        if isinstance(billed, (int, float)):
+            parts.append(f"billed ${billed:.6f}".rstrip("0").rstrip("."))
+
+        # DELIBERATELY NOT rendering `savesVsDirect` as a percentage saved.
+        # Measured: billedUsd 0.00700472 / wouldHaveBilled 0.02101416 =
+        # 0.3333 == savesVsDirect. So the field is the RATIO PAID, not the
+        # fraction saved -- printing "saved 0.33" claims a third when the
+        # real saving is two thirds. Show both absolute numbers instead;
+        # they cannot be misread.
+        would = sub.get("wouldHaveBilled") or x.get("directUsd")
+        if isinstance(would, (int, float)) and isinstance(billed, (int, float)) \
+                and would > billed > 0:
+            parts.append(f"vs ${would:.6f}".rstrip("0").rstrip(".") + " direct"
+                         f" ({would / billed:.1f}x)")
+
+        if sub.get("tier"):
+            parts.append(f"{sub['tier']} subscription")
+        elif x.get("paid"):
+            parts.append(str(x["paid"]))
+
+        pt = usage.get("prompt_tokens")
+        if pt:
+            parts.append(f"{pt} tokens read")
+        return ("\n\n---\n*openzoo: " + " · ".join(parts) + "*") if parts else ""
 
     # ------------------------------------------------------------------ pipes
 
@@ -186,8 +221,7 @@ def pipe(self, body: dict):
 
         data = r.json()
         if self.valves.SHOW_RECEIPTS:
-            usage = data.get("usage", {}) or {}
-            receipt = self._format_receipt(usage)
+            receipt = self._format_receipt(data)   # whole body: x402 is top-level
             if receipt:
                 try:
                     data["choices"][0]["message"]["content"] += receipt
@@ -217,9 +251,27 @@ def _selftest():
             "chat_id": "x", "session_id": "y", "metadata": {}, "features": {}}
     clean = {k: v for k, v in junk.items() if k in Pipe._OPENAI_CHAT_FIELDS}
     assert set(clean) == {"model", "messages", "stream"}, clean
-    # receipt formatting: all fields, some fields, none
-    full = Pipe._format_receipt({"billedUsd": 0.0021, "savesVsDirect": "97%", "prompt_tokens": 3100})
-    assert "billed $0.0021" in full and "saved 97%" in full and "3100 tokens read" in full
+    # receipt formatting -- REGRESSION TEST for the bug this file used to have.
+    # These are the exact shapes a live gateway returns; the old code read
+    # `usage` for billedUsd/savesVsDirect and therefore rendered nothing.
+    live = {
+        "usage": {"prompt_tokens": 3100, "completion_tokens": 20, "cost": 0.007},
+        "x402": {"billedUsd": 0.00700472, "savesVsDirect": 0.3333333333333333,
+                 "cogsUsd": 0.00700472, "directUsd": 0.00700472,
+                 "paid": "subscription",
+                 "subscription": {"tier": "pro", "wouldHaveBilled": 0.02101416,
+                                  "invoiced": "stripe"}},
+    }
+    full = Pipe._format_receipt(live)
+    assert "billed $0.007" in full, full
+    assert "vs $0.021" in full and "3.0x" in full, full   # 0.021/0.007
+    assert "pro subscription" in full, full
+    assert "3100 tokens read" in full, full
+    # and the ratio must NOT be printed as a savings percentage
+    assert "saved" not in full, full
+    # a usage-only body (no x402 -- e.g. prepaid credit, no 402 emitted) still
+    # renders the token count rather than blowing up
+    assert "3100 tokens read" in Pipe._format_receipt({"usage": {"prompt_tokens": 3100}})
     assert Pipe._format_receipt({}) == ""
     # valves defaults are the local-proxy contract
     assert p.valves.OPENZOO_BASE_URL.endswith("/v1")

integrations/README.md

@@ -62,7 +62,7 @@ Every integration here targets one or both of openzoo's surfaces:
    short prompts price at a 3× passthrough markup — receipts name which base
    applied, and print on the proxy console per call.
 2. **MCP server** — `npx openzoo mcp` (stdio). Tools: `zoo_ask` (corpus up to
-   ~1M tokens + question → answer + receipt), `zoo_models`, `zoo_wallet`. MCP
+   ~9.8M tokens per call (~128M bound ceiling) + question → answer + receipt), `zoo_models`, `zoo_wallet`. MCP
    hosts (Cursor, Cline, Claude Desktop, Windsurf) should wire BOTH surfaces —
    chat for ordinary completions, MCP for the giant-corpus flagship.
 

integrations/SillyTavern/README.md

@@ -11,7 +11,7 @@ configuration — there is no file to install, so this folder is documentation o
 3. Set **API** → `Chat Completion`.
 4. Set **Chat Completion Source** → `Custom (OpenAI-compatible)`.
 5. **Custom Endpoint**: `http://localhost:8402/v1`
-   (hosted: `https://api.openzoo.fun/v1`)
+   (hosted: `https://x402-tokens.fly.dev/v1`)
    — stop at `/v1`; do NOT append `/chat/completions`, SillyTavern adds the route.
 6. **API Key**: `sk-openzoo` (any value for the local proxy; real key for hosted).
 7. Connect. Because the proxy implements `/v1/models`, the model dropdown fills

integrations/aider/README.md

@@ -11,7 +11,7 @@ source integrations/aider/openzoo.env          # sets the two variables
 aider --model openai/nvidia/nemotron-3.5-lightning         # any zoo model id works
 ```
 
-`npx openzoo models` lists model ids and pricing.
+`curl localhost:8402/v1/models` lists model ids and pricing.
 
 ## Persistent config (optional)
 

integrations/aider/openzoo.env

@@ -9,5 +9,5 @@
 # openai/-prefixed model as living at that base. Two exports is the whole
 # integration.
 
-export OPENAI_API_BASE="http://localhost:8402/v1"   # hosted: https://api.openzoo.fun/v1
+export OPENAI_API_BASE="http://localhost:8402/v1"   # hosted: https://x402-tokens.fly.dev/v1
 export OPENAI_API_KEY="sk-openzoo"                  # any value for local proxy; real key for hosted