Skip to content

Free, open-source AI video generator for Mac. Run LTX-2.5 and MiniMax H3 (Hailuo) locally on Apple Silicon: text-to-video and image-to-video with sound, lip-sync, character LoRA training, songs from lyrics (YuE2) and image editing (Qwen-Image-Edit). One-click install via Pinokio. Offline: no cloud, no API key, no NVIDIA GPU.

revert(analytics): remove opt-in telemetry entirely

mrbizarroda1d6f52bc8c
diff --git a/README.md b/README.md
index 5b16097..4e27104 100644
--- a/README.md
+++ b/README.md
@@ -15,7 +15,7 @@
## Overview
-Phosphene is a local generative-media panel for Apple Silicon. It runs [LTX-Video 2.3](https://github.com/Lightricks/LTX-Video) (MLX port) for joint audio-and-video synthesis, [Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511) and an MLX port of [HiDream-O1-Image-Dev](https://huggingface.co/HiDream-ai/HiDream-O1-Image-Dev) for stills, and ships an in-panel LoRA training pipeline for character identity (face + optional voice from a single dataset). Everything runs on-device. No cloud, no API keys. Anonymous usage analytics are opt-in and OFF by default — see [TELEMETRY.md](TELEMETRY.md).
+Phosphene is a local generative-media panel for Apple Silicon. It runs [LTX-Video 2.3](https://github.com/Lightricks/LTX-Video) (MLX port) for joint audio-and-video synthesis, [Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511) and an MLX port of [HiDream-O1-Image-Dev](https://huggingface.co/HiDream-ai/HiDream-O1-Image-Dev) for stills, and ships an in-panel LoRA training pipeline for character identity (face + optional voice from a single dataset). Everything runs on-device. No cloud, no API keys, no telemetry.
3.0 introduces in-panel character training (face + voice LoRA from one dataset), the Audio-to-Video workflow, the Image Studio tab with two MLX-native engines, hardware capability tiering, and an agentic prompt enhancer / shot planner driven by the same local Gemma 3 12B used for auto-captioning.
@@ -248,6 +248,4 @@ Phosphene is free and open source.
## Network note
-Phosphene runs locally. A clean production install checks GitHub every 30 minutes for an update badge, and only touches Hugging Face or CivitAI when you download models or LoRAs. Disable the update check with `PHOSPHENE_DISABLE_VERSION_CHECK=1`. The panel binds to `127.0.0.1` with no auth. It's not designed for LAN exposure or tunneling.
-
-**Telemetry.** OFF by default. If you opt in under **Settings → Anonymous analytics**, the panel sends a small JSON event per render — type, duration, hardware tier, error category. Never prompts, image bytes, filenames, paths, or hostname. Full schema and rationale: [TELEMETRY.md](TELEMETRY.md). Point at a self-hosted endpoint with `PHOSPHENE_ANALYTICS_ENDPOINT`.
+Phosphene runs locally. No telemetry. A clean production install checks GitHub every 30 minutes for an update badge, and only touches Hugging Face or CivitAI when you download models or LoRAs. Disable the update check with `PHOSPHENE_DISABLE_VERSION_CHECK=1`. The panel binds to `127.0.0.1` with no auth. It's not designed for LAN exposure or tunneling.
diff --git a/TELEMETRY.md b/TELEMETRY.md
deleted file mode 100644
index efd38be..0000000
--- a/TELEMETRY.md
+++ /dev/null
@@ -1,182 +0,0 @@
-# Phosphene anonymous telemetry
-
-## TL;DR
-
-- **OFF by default.** Enable it in **Settings → Anonymous analytics**.
-- Sends a small JSON event per render boundary. No prompts. No image
- bytes. No filenames. No paths. No hostname. No IP-in-payload.
-- A single anonymous install id (32-char hex) is generated the first
- time you opt in. Click **forget me** to rotate it.
-- All HTTP is non-blocking and best-effort. A flaky endpoint will never
- stall a render.
-- Endpoint configurable via `PHOSPHENE_ANALYTICS_ENDPOINT` env var.
- Default: `https://analytics.phosphene.ai/v1/event` (run by Mr Bizarro).
-
-## Why
-
-Phosphene is a young open-source project. GitHub release-download stats
-only show a 14-day window, and Pinokio installs aren't counted at all.
-Without **opt-in** telemetry we have no way to tell:
-
-- which Apple Silicon tiers people actually run on (M2, M3, M4 Max,
- Ultra…) so we can prioritize the right perf wins;
-- which modes (T2V, I2V, A2V, FFLF, Extend, Character) get used most
- vs. which are gathering dust;
-- what's breaking in the field — OOMs, missing models, helper crashes,
- bad LoRA combos — so we can ship fixes rather than wait for an issue
- to land;
-- whether a release regressed render time on real hardware.
-
-If you'd rather not share any of this, leave the toggle OFF — Phosphene
-works identically. We mean it.
-
-## What we collect
-
-Every event carries:
-
-| Field | Example | Notes |
-|---------------|----------------------|------------------------------------------|
-| `event` | `render_done` | The event type — see list below. |
-| `ts` | `2026-05-21T19:42:11Z` | ISO 8601 UTC, second precision. |
-| `install_id` | `a1b2c3…` | Anonymous 32-char hex. Generated locally on first opt-in; you can rotate it. |
-
-Plus, **per event type**:
-
-### `panel_boot` — fires once at panel startup
-
-| Field | Example | Why |
-|-----------------|------------------------|--------------------------------------|
-| `version` | `0.10.3` | Phosphene release tag if known. |
-| `sha` | `ec1437f` | Short commit if you're on dev. |
-| `branch` | `dev` | `main` or `dev`. |
-| `tier` | `M4 Max 64 GB` | Hardware tier label. |
-| `allows_q8` | `true` | Whether the machine can run Q8 HQ. |
-| `os` | `macOS 26.4` | OS family + major.minor only. |
-| `machine` | `arm64` | CPU architecture. |
-| `python` | `3.11` | Major.minor. |
-| `ram_gb_bucket` | `64` | Snapped to nearest bucket (16/32/48/64/96/128/192/256). Raw GB would near-uniquely identify users. |
-| `cpu_brand` | `Apple M4 Max` | Truncated to 64 chars. |
-
-### `render_start` / `render_done` / `render_failed` / `render_cancelled`
-
-| Field | Example | Notes |
-|------------------|---------------|---------------------------------------------|
-| `mode` | `i2v` | One of: t2v, i2v, a2v, character, fflf, extend, image, audio. |
-| `quality` | `q8_hq` | Pipeline tier picked. |
-| `width` | `1280` | Output width in pixels. |
-| `height` | `704` | Output height. |
-| `frames` | `97` | Frame count requested. |
-| `has_character` | `true` | Whether a Phosphene character LoRA is loaded — not which one. |
-| `lora_count` | `2` | Number of user LoRAs (count only, no names).|
-| `hdr` | `false` | HDR IC-LoRA pipeline? |
-| `enhance` | `true` | Gemma prompt-enhance step run? |
-| `upscale` | `false` | Post-upscale step run? |
-| `accel` | `exact` | Fast / Exact pill. |
-| `engine` | `mlx` | Pipeline engine (mlx / comfy). |
-| `elapsed_sec` | `306.5` | End-to-end seconds. Only on `_done`/`_failed`/`_cancelled`. |
-| `error_category` | `oom` | Only on `_failed`. Coarse bucket — see list below. |
-
-Error categories: `oom`, `hf_auth`, `hf_gated`, `wrong_backend`,
-`missing_model`, `q8_missing`, `missing_deps`, `character_error`,
-`file_missing`, `helper_pipe`, `runtime`.
-
-### `helper_crash` — fires when the helper subprocess exits unexpectedly
-
-| Field | Example | Notes |
-|-----------------|--------------|----------------------------------------|
-| `exit_code` | `-9` | OS exit code. |
-| `last_action` | `generate` | What the helper was doing. |
-| `mode` | `i2v` | Same shape as above. |
-
-### `settings_opt_in` / `settings_opt_out`
-
-| Field | Example | Notes |
-|-----------|---------|---------------------------------------------|
-| `version` | `0.10.3`| Phosphene release tag at the moment of the flip. |
-
-`settings_opt_out` is the **last event** sent before silence. After
-that we send nothing until you opt in again.
-
-## What we **never** collect
-
-The wire sanitizer drops any field with these keys, even if a caller
-accidentally tries to attach one:
-
-```
-prompt, negative_prompt, image, image_path, audio_path,
-output, output_path, path, filename,
-hostname, username, user
-```
-
-Beyond that:
-
-- **No URL parameters** — events are POSTed as JSON bodies.
-- **No IP in payload** — your IP is visible to the receiving server at
- the TCP layer (same as any HTTP request) but it's not stored or
- attached to the install_id.
-- **No reference image bytes / no rendered frames / no audio.**
-- **No GitHub username, no machine name, no email, no API tokens.**
-- **No LoRA filenames** — we only count how many you used.
-
-## Defense in depth
-
-Even if a future code change tried to ship more data than the schema
-above, `analytics.py` enforces hard limits:
-
-- **4 KB max payload.** Anything over gets dropped, not truncated.
-- **200-char max per string value.** Runaway error strings get cut.
-- **512-event in-memory queue.** When the endpoint is unreachable,
- oldest events drop first — never unbounded growth.
-- **4 s socket timeout.** A flaky endpoint can't stall a render.
-- **Daemon thread.** The flusher dies with the process; it never holds
- shutdown open.
-
-The full implementation lives in `analytics.py` — 350 lines, readable
-top-to-bottom in under 10 minutes. Audit before you opt in.
-
-## Controlling it
-
-### Opt in / opt out
-
-**Settings → Anonymous analytics → Enable / Disable.**
-
-Takes effect immediately. No restart, no helper kick.
-
-### Rotate your anonymous id ("forget me")
-
-**Settings → Anonymous analytics → forget me** (button next to the id).
-
-Generates a fresh 32-char hex. Past events from your old id stay in
-the receiver's logs but can no longer be correlated to your new
-events.
-
-### Self-host the endpoint
-
-Set `PHOSPHENE_ANALYTICS_ENDPOINT` in the panel's environment:
-
-```bash
-export PHOSPHENE_ANALYTICS_ENDPOINT="https://my-receiver.example.com/event"
-```
-
-If both the env var and the saved endpoint are unset, telemetry is a
-no-op even when the toggle is ON (fail-safe for forks).
-
-### Disable entirely at build time
-
-Leave the toggle OFF. Done. The code path is a no-op — `emit()` short-
-circuits inside `Analytics.emit` when `is_enabled()` returns False.
-
-## Receiver
-
-The default endpoint at `analytics.phosphene.ai` accepts POST JSON,
-returns `204 No Content`, retains events for 90 days, and drops any
-field outside the schema above on ingest. Source for the receiver is
-not yet public — happy to publish it once the schema stabilizes. File
-an issue if you'd like to see it sooner.
-
-## Questions?
-
-Open an issue at <https://github.com/mrbizarro/phosphene/issues> with
-the `telemetry` label. The privacy posture is intentionally
-conservative — if any of it makes you uncomfortable, please tell us so
-we can fix it.
diff --git a/analytics.py b/analytics.py
deleted file mode 100644
index d59e408..0000000
--- a/analytics.py
+++ /dev/null
@@ -1,359 +0,0 @@
-"""Phosphene anonymous telemetry.
-
-Strictly opt-in. Default OFF. When the user enables it via Settings,
-the panel sends a tiny event payload to a configurable endpoint
-(``PHOSPHENE_ANALYTICS_ENDPOINT`` env var, falls back to the public
-ingest URL Mr Bizarro hosts).
-
-Privacy guarantees:
-- Never sends prompts, image bytes, output filenames, audio waveforms,
- LoRA filenames (only counts), absolute paths, hostnames, IP, or any
- user-content.
-- Anonymous install UUID generated once on first opt-in. Rotating
- ``analytics_install_id`` in settings resets identity (forget-me).
-- All HTTP calls are non-blocking (daemon thread, 4 s socket timeout) so
- a slow endpoint never stalls a render.
-- Endpoint failures are logged to the panel log but never raised — the
- user's render is the priority, telemetry is best-effort.
-
-Full event schema:
- See TELEMETRY.md alongside this file.
-
-Events emitted today:
- panel_boot — every panel startup (version, branch, tier, RAM)
- render_start — every job enters the worker
- render_done — every successful job (with elapsed_sec, mode, quality)
- render_failed — every error / cancellation (with categorized error)
- helper_crash — every unexpected helper subprocess exit
- settings_opt_in — when the user flips analytics_enabled on
- settings_opt_out — when the user flips analytics_enabled off (last
- event before silence)
-"""
-from __future__ import annotations
-
-import json
-import os
-import platform
-import queue
-import sys
-import threading
-import time
-import urllib.error
-import urllib.request
-import uuid
-from typing import Any
-
-
-# Default public ingest. Mr Bizarro can swap this via env. If neither
-# env nor saved settings provide an endpoint, telemetry is a no-op
-# even when analytics_enabled=True — fail-safe for forks.
-DEFAULT_ENDPOINT = os.environ.get(
- "PHOSPHENE_ANALYTICS_ENDPOINT",
- "https://analytics.phosphene.ai/v1/event",
-).strip()
-
-# Hard limit on payload size — defense-in-depth. Any single event over
-# this gets dropped without sending (prevents accidental large blobs).
-_MAX_PAYLOAD_BYTES = 4 * 1024
-
-# Hard limit on the in-memory event queue. Older events get dropped if
-# the endpoint is unreachable and events pile up. Prevents unbounded
-# memory growth across a long session.
-_QUEUE_MAX = 512
-
-# Socket timeout per request. Keep tight so a flaky endpoint never
-# stalls the render path.
-_SOCKET_TIMEOUT_SEC = 4
-
-# How long to wait between background-flush ticks. Most events are
-# already triggered by user actions (render boundaries), so the
-# background loop is really just retry / drain.
-_FLUSH_INTERVAL_SEC = 5
-
-
-# Module-level singleton — the ``Analytics`` class is a thread that
-# owns the event queue + background flusher. We construct it once at
-# panel startup and write through the module-level helpers below.
-_INSTANCE: "Analytics | None" = None
-_INSTANCE_LOCK = threading.Lock()
-
-
-def _now_iso() -> str:
- """ISO 8601 UTC timestamp, second precision (sub-second isn't useful
- for cross-machine event correlation and adds noise)."""
- return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
-
-
-class Analytics:
- """Background-thread event sender.
-
- Construct once at panel startup. The constructor does NOT start the
- thread — call ``start()`` after the settings layer is wired up so
- we can read the opt-in flag.
- """
-
- def __init__(
- self,
- *,
- is_enabled: callable,
- get_install_id: callable,
- get_endpoint: callable,
- log: callable | None = None,
- ) -> None:
- self._is_enabled = is_enabled
- self._get_install_id = get_install_id
- self._get_endpoint = get_endpoint
- # ``log`` should be the panel's ``push()`` (or any function that
- # accepts a single string). When None, errors go to stderr only.
- self._log = log or (lambda s: sys.stderr.write(s + "\n"))
- self._queue: "queue.Queue[dict[str, Any]]" = queue.Queue(maxsize=_QUEUE_MAX)
- self._stop_event = threading.Event()
- self._thread: threading.Thread | None = None
-
- # ------------------------------------------------------------------
- # Thread lifecycle
- # ------------------------------------------------------------------
-
- def start(self) -> None:
- """Start the background flush thread. Idempotent."""
- if self._thread and self._thread.is_alive():
- return
- self._stop_event.clear()
- self._thread = threading.Thread(
- target=self._run, name="phosphene-analytics", daemon=True
- )
- self._thread.start()
-
- def stop(self) -> None:
- """Signal the background thread to exit and join briefly. Called
- at process shutdown (best-effort — daemon thread will die with
- the process anyway)."""
- self._stop_event.set()
- if self._thread:
- self._thread.join(timeout=2.0)
-
- # ------------------------------------------------------------------
- # Event entry points
- # ------------------------------------------------------------------
-
- def emit(self, event_type: str, fields: dict[str, Any] | None = None) -> None:
- """Queue an event. No-op if analytics is disabled. Safe to call
- from any thread; never raises."""
- try:
- if not self._is_enabled():
- return
- payload = self._build_event(event_type, fields or {})
- if payload is None:
- return # event was dropped by validation
- try:
- self._queue.put_nowait(payload)
- except queue.Full:
- # Drop the OLDEST item, queue the new one. Recent events
- # are more useful than ancient ones when we're behind.
- try:
- self._queue.get_nowait()
- except queue.Empty:
- pass
- try:
- self._queue.put_nowait(payload)
- except queue.Full:
- pass
- except Exception as exc: # noqa: BLE001 — never crash the caller
- try:
- self._log(f"analytics: emit failed silently: {exc}")
- except Exception:
- pass
-
- # ------------------------------------------------------------------
- # Internal
- # ------------------------------------------------------------------
-
- def _build_event(self, event_type: str, fields: dict[str, Any]) -> dict[str, Any] | None:
- """Compose the final event dict + run privacy validation.
-
- Returns ``None`` if the event should be silently dropped (e.g.
- oversize payload or a forbidden field name leaked in).
- """
- # Strip any fields whose key looks like it could carry user
- # content. Defense in depth — even if a caller violates the
- # rule, the sanitizer keeps the wire clean.
- BANNED = {"prompt", "image", "image_path", "audio_path",
- "output_path", "output", "path", "hostname",
- "username", "user", "negative_prompt", "filename"}
- clean: dict[str, Any] = {}
- for k, v in fields.items():
- if k in BANNED:
- continue
- # Truncate stringy values to 200 chars. Some fields are
- # legitimately stringy (e.g., error category) but a runaway
- # value should never make it to wire.
- if isinstance(v, str) and len(v) > 200:
- v = v[:200] + "…"
- clean[k] = v
- event = {
- "event": str(event_type)[:64],
- "ts": _now_iso(),
- "install_id": self._get_install_id(),
- **clean,
- }
- # Size check
- try:
- payload_bytes = json.dumps(event).encode("utf-8")
- except (TypeError, ValueError) as exc:
- self._log(f"analytics: event {event_type!r} not JSON-serializable: {exc}")
- return None
- if len(payload_bytes) > _MAX_PAYLOAD_BYTES:
- self._log(
- f"analytics: dropping oversize {event_type} event "
- f"({len(payload_bytes)} > {_MAX_PAYLOAD_BYTES} bytes)"
- )
- return None
- return event
-
- def _run(self) -> None:
- """Background loop. Drains the queue, POSTs each event with a
- short timeout. Failures are logged once per minute (silent
- otherwise) to avoid log spam from a flaky endpoint."""
- last_warn_ts = 0.0
- while not self._stop_event.is_set():
- try:
- event = self._queue.get(timeout=_FLUSH_INTERVAL_SEC)
- except queue.Empty:
- continue
- try:
- self._post(event)
- except Exception as exc: # noqa: BLE001
- now = time.time()
- if now - last_warn_ts > 60:
- last_warn_ts = now
- try:
- self._log(f"analytics: post failed ({exc}); will retry")
- except Exception:
- pass
- # Put it back? Decision: NO — accept the drop. Otherwise
- # a permanently-down endpoint makes us cycle forever.
- # The settings_opt_out event is the one drop we'd hate
- # to lose; we accept that loss for simplicity.
-
- def _post(self, event: dict[str, Any]) -> None:
- endpoint = self._get_endpoint()
- if not endpoint:
- return # no endpoint configured; silent no-op
- body = json.dumps(event).encode("utf-8")
- req = urllib.request.Request(
- endpoint,
- data=body,
- method="POST",
- headers={
- "Content-Type": "application/json",
- "User-Agent": "phosphene-panel-analytics",
- },
- )
- with urllib.request.urlopen(req, timeout=_SOCKET_TIMEOUT_SEC) as resp:
- # We don't care about the response body; just consume so
- # the connection can close cleanly.
- try:
- resp.read(64)
- except Exception: # noqa: BLE001
- pass
-
-
-# ---------------------------------------------------------------------
-# Module-level helpers — the panel calls these directly
-# ---------------------------------------------------------------------
-
-def install(
- *,
- is_enabled: callable,
- get_install_id: callable,
- get_endpoint: callable,
- log: callable | None = None,
-) -> None:
- """Wire up the singleton. Call once at panel boot before any emit().
- Idempotent — subsequent calls are no-ops."""
- global _INSTANCE
- with _INSTANCE_LOCK:
- if _INSTANCE is not None:
- return
- _INSTANCE = Analytics(
- is_enabled=is_enabled,
- get_install_id=get_install_id,
- get_endpoint=get_endpoint,
- log=log,
- )
- _INSTANCE.start()
-
-
-def emit(event_type: str, fields: dict[str, Any] | None = None) -> None:
- """Queue an event. No-op if analytics is disabled or not installed."""
- inst = _INSTANCE
- if inst is not None:
- inst.emit(event_type, fields)
-
-
-def shutdown() -> None:
- """Stop the background thread. Best-effort; daemon thread dies
- with the process anyway."""
- inst = _INSTANCE
- if inst is not None:
- inst.stop()
-
-
-def new_install_id() -> str:
- """Generate a fresh anonymous install UUID. Called by the settings
- layer the first time analytics_enabled flips ON (or when the user
- rotates the id via the 'forget me' button)."""
- return uuid.uuid4().hex
-
-
-def system_fingerprint() -> dict[str, Any]:
- """Anonymized hardware / software fingerprint. Sent with panel_boot
- so we can correlate bugs to hardware tier + OS version without
- learning anything user-identifying.
-
- What's collected:
- os — "macOS 26.4" etc., no patch level
- machine — "arm64"
- python — "3.11"
- ram_gb_bucket — bucketed: 16/32/48/64/96/128/192/256+
- (raw value would near-uniquely identify
- a user across the install pool)
- cpu_brand — "Apple M4 Max" etc. when available
- """
- info: dict[str, Any] = {}
- try:
- sys_name = platform.system()
- if sys_name == "Darwin":
- # "26.4.1" → "26.4" — patch level doesn't add bug-correlation
- # value and slightly increases per-user uniqueness.
- mac = (platform.mac_ver()[0] or "").split(".")
- major_minor = ".".join(mac[:2]) if mac and mac[0] else ""
- info["os"] = f"macOS {major_minor}".strip()
- else:
- info["os"] = sys_name or "?"
- except Exception:
- info["os"] = "?"
- info["machine"] = platform.machine() or "?"
- info["python"] = ".".join(platform.python_version().split(".")[:2])
- try:
- # psutil is already a Phosphene dep
- import psutil # noqa: PLC0415
- ram_gb = psutil.virtual_memory().total / (1024**3)
- for bucket in (16, 32, 48, 64, 96, 128, 192, 256):
- if ram_gb <= bucket + 4: # snap to bucket within ±4 GB
- info["ram_gb_bucket"] = bucket
- break
- else:
- info["ram_gb_bucket"] = 256
- except Exception:
- info["ram_gb_bucket"] = 0
- try:
- import subprocess # noqa: PLC0415
- out = subprocess.check_output(
- ["sysctl", "-n", "machdep.cpu.brand_string"],
- timeout=2, stderr=subprocess.DEVNULL,
- ).decode().strip()
- info["cpu_brand"] = out[:64]
- except Exception:
- info["cpu_brand"] = "?"
- return info
diff --git a/mlx_ltx_panel.py b/mlx_ltx_panel.py
index f1055f1..216e115 100644
--- a/mlx_ltx_panel.py
+++ b/mlx_ltx_panel.py
@@ -334,21 +334,13 @@ def _settings_defaults() -> dict:
# Memory/speed policy. Defaults to Auto so 5 s clips keep the fast
# full-decode path while long/high-pressure renders stay protected.
"memory_policy": DEFAULT_MEMORY_POLICY,
- # Anonymous telemetry — OPT-IN. Default OFF. When enabled, the
- # panel sends a tiny event payload (no prompts, no image bytes,
- # no paths, no hostnames) for every render start/done/failed, the
- # panel boot itself, and helper crashes. Used to understand
- # version regressions, hardware-tier feature usage, and bug
- # frequency over time — github clone-counts only retain 14 days.
- # Full event schema lives in TELEMETRY.md alongside this file.
- # Endpoint configurable via PHOSPHENE_ANALYTICS_ENDPOINT env var.
- "analytics_enabled": False,
- # Anonymous install UUID — generated once on first opt-in and
- # written here. Lets us count unique installs without correlating
- # with any user identity. Rotating this field resets the install
- # identity (useful if a user wants to "forget me" before
- # re-enabling).
- "analytics_install_id": "",
+ # NOTE: opt-in anonymous telemetry was experimented with in
+ # 2026-05 and removed on 2026-05-22 (see commit history) — Mr
+ # Bizarro chose to stay on GitHub-data-only signal (see
+ # docs/stats.html + scripts/fetch_repo_stats.py) rather than
+ # ship anything that touches the user's render path. Existing
+ # panel_settings.json files may still have analytics_* keys;
+ # they're now silently ignored.
}
@@ -562,30 +554,9 @@ def _validate_settings_patch(patch: dict) -> tuple[dict, str | None]:
return {}, f"unknown memory_policy: {policy}"
out["memory_policy"] = policy
- if "analytics_enabled" in patch:
- # urlencoded-bool coercion. When flipping ON, generate a fresh
- # anonymous install_id if one isn't already saved — that's the
- # only stable identifier we ever attach to events. Rotating it
- # via the panel "Forget me" button (which clears
- # analytics_install_id) is how the user resets identity.
- v = patch["analytics_enabled"]
- if isinstance(v, bool):
- new_val = v
- else:
- new_val = str(v).strip().lower() in ("1", "true", "yes", "on")
- out["analytics_enabled"] = new_val
-
- if "analytics_install_id" in patch:
- v = str(patch["analytics_install_id"]).strip()
- # Allow only an empty string (forget-me) OR a 32-char hex UUID.
- # Anything else is rejected to avoid the field becoming a free-
- # form data sink the user could put PII in.
- if v == "":
- out["analytics_install_id"] = ""
- elif len(v) == 32 and all(c in "0123456789abcdefABCDEF" for c in v):
- out["analytics_install_id"] = v.lower()
- else:
- return {}, "analytics_install_id must be empty or 32-char hex"
+ # analytics_enabled / analytics_install_id used to be validated here;
+ # both removed 2026-05-22 when opt-in analytics was rolled back. Any
+ # legacy keys in panel_settings.json are now silently ignored.
return out, None
@@ -630,11 +601,6 @@ def get_settings_public() -> dict:
"models_card_dismissed": bool(s.get("models_card_dismissed", False)),
"spicy_mode": bool(s.get("spicy_mode", False)),
"memory_policy": s.get("memory_policy", DEFAULT_MEMORY_POLICY),
- # Anonymous opt-in telemetry. The install_id itself is fine to
- # expose — it's not a secret, the user can rotate it from the
- # UI ("forget me"). The flag drives the Settings UI toggle.
- "analytics_enabled": bool(s.get("analytics_enabled", False)),
- "analytics_install_id": str(s.get("analytics_install_id", "")),
}
@@ -7085,28 +7051,6 @@ def worker_loop() -> None:
STATE["log"] = []
caffeinate_on()
persist_queue()
- # ANALYTICS — render_start. Sanitized: no prompt, no paths, no
- # filenames. Just shape (mode/quality/dims/frames). Safe-call:
- # any failure inside emit() is swallowed silently.
- try:
- import analytics as _ph_analytics # type: ignore
- _p = job.get("params", {}) or {}
- _ph_analytics.emit("render_start", {
- "mode": _p.get("mode"),
- "quality": _p.get("quality"),
- "width": _p.get("width"),
- "height": _p.get("height"),
- "frames": _p.get("frames"),
- "has_character": bool(_p.get("character_id")),
- "lora_count": len(_p.get("loras") or []),
- "hdr": bool(_p.get("hdr")),
- "enhance": bool(_p.get("enhance")),
- "upscale": _p.get("upscale"),
- "accel": _p.get("accel"),
- "engine": _p.get("engine_override") if _p.get("mode") == "image" else None,
- })
- except Exception: # noqa: BLE001
- pass
try:
run_job_inner(job)
@@ -7131,53 +7075,6 @@ def worker_loop() -> None:
STATE["pid"] = None
STATE["pgid"] = None
persist_queue()
- # ANALYTICS — render_done / render_failed / render_cancelled.
- # The error_category bucketing is coarse on purpose: keep the
- # cardinality low so the same kind of bug across users
- # aggregates cleanly. Full error STRINGS would leak filesystem
- # paths and prompt fragments.
- try:
- import analytics as _ph_analytics # type: ignore
- _p = job.get("params", {}) or {}
- _status = job.get("status", "?")
- _ev = {
- "done": "render_done",
- "failed": "render_failed",
- "cancelled": "render_cancelled",
- }.get(_status, "render_unknown")
- _evt: dict = {
- "mode": _p.get("mode"),
- "quality": _p.get("quality"),
- "elapsed_sec": job.get("elapsed_sec"),
- "frames": _p.get("frames"),
- }
- if _status == "failed":
- # Categorize the error so the wire stays small + private.
- _err = (job.get("error") or "")
- _cat = "other"
- for needle, name in (
- ("out of memory", "oom"),
- ("OOM", "oom"),
- ("jetsam", "oom"),
- ("HF", "hf_auth"),
- ("gated", "hf_gated"),
- ("401", "hf_auth"),
- ("CUDA", "wrong_backend"),
- ("model_dir", "missing_model"),
- ("Q8", "q8_missing"),
- ("not installed", "missing_deps"),
- ("character", "character_error"),
- ("FileNotFoundError", "file_missing"),
- ("BrokenPipeError", "helper_pipe"),
- ("RuntimeError", "runtime"),
- ):
- if needle.lower() in _err.lower():
- _cat = name
- break
- _evt["error_category"] = _cat
- _ph_analytics.emit(_ev, _evt)
- except Exception: # noqa: BLE001
- pass
# ---- Image Studio + characters integration ----------------------------------
@@ -11061,56 +10958,9 @@ class Handler(BaseHTTPRequestHandler):
prev.get("memory_policy", DEFAULT_MEMORY_POLICY) !=
current.get("memory_policy", DEFAULT_MEMORY_POLICY)
)
- # Analytics opt-in lifecycle. When flipping ON, generate a
- # fresh anonymous install_id if one isn't already saved, and
- # fire a single settings_opt_in event so we count adoption.
- # When flipping OFF, fire settings_opt_out as the LAST event
- # before silence (gives us the offboarding signal without
- # tracking the user afterward).
- analytics_was = bool(prev.get("analytics_enabled", False))
- analytics_now = bool(current.get("analytics_enabled", False))
- if analytics_now and not analytics_was:
- if not current.get("analytics_install_id"):
- try:
- import analytics as _ph_a # type: ignore
- new_id = _ph_a.new_install_id()
- update_settings({"analytics_install_id": new_id})
- current = get_settings()
- push("settings: analytics ON (anonymous install id generated)")
- except Exception as _e: # noqa: BLE001
- push(f"settings: analytics ON but install_id gen failed: {_e}")
- try:
- import analytics as _ph_a # type: ignore
- _ph_a.emit("settings_opt_in", {
- "version": _read_local_version() or "unknown",
- })
- except Exception: # noqa: BLE001
- pass
- elif analytics_was and not analytics_now:
- # Last gasp before silence.
- try:
- import analytics as _ph_a # type: ignore
- _ph_a.emit("settings_opt_out", {
- "version": _read_local_version() or "unknown",
- })
- except Exception: # noqa: BLE001
- pass
- elif analytics_now and analytics_was:
- # Forget-me path — user blanked install_id while still
- # opted in. Regenerate immediately so the next event
- # uses a fresh anonymous identity (anyone correlating
- # events on the ingest side sees this as a new user).
- prev_id = (prev.get("analytics_install_id") or "").strip()
- cur_id = (current.get("analytics_install_id") or "").strip()
- if prev_id and not cur_id:
- try:
- import analytics as _ph_a # type: ignore
- new_id = _ph_a.new_install_id()
- update_settings({"analytics_install_id": new_id})
- current = get_settings()
- push("settings: analytics identity rotated")
- except Exception as _e: # noqa: BLE001
- push(f"settings: analytics rotate failed: {_e}")
+ # Analytics opt-in lifecycle removed 2026-05-22 (the entire
+ # analytics surface was rolled back — we stand on GitHub-data
+ # signal via docs/stats.html instead).
if codec_changed:
push(
f"settings: output codec → {current['output_pix_fmt']} "
@@ -19562,36 +19412,6 @@ HTML = r"""<!doctype html>
<div class="hint" id="spicyHint" style="margin-top:8px; display:none"></div>
</div>
- <!-- Anonymous opt-in telemetry. Strictly OFF by default. When ON,
- a tiny event payload (no prompts, no image bytes, no paths) is
- sent per render so we can spot bugs + understand how Phosphene
- is actually used. The "What's collected?" link opens the
- TELEMETRY.md doc shipping in the repo so the user can audit
- every field before opting in. -->
- <div class="settings-section">
- <h3>Anonymous analytics <span class="hint" style="font-weight:400">(help improve Phosphene)</span></h3>
- <div class="hint" style="margin-bottom:10px">
- OFF by default. When ON, Phosphene sends a tiny anonymous event
- per render — render type, duration, model tier, error category.
- Never sends prompts, image data, file paths, or your hostname.
- <a href="https://github.com/mrbizarro/phosphene/blob/main/TELEMETRY.md"
- target="_blank" rel="noopener">What's collected?</a>
- </div>
- <div class="spicy-row">
- <span class="spicy-state" id="analyticsStateBadge">OFF</span>
- <button type="button" class="ghost-btn" id="analyticsToggleBtn"
- onclick="toggleAnalytics()" style="margin-left:auto">
- Enable anonymous analytics
- </button>
- </div>
- <div class="hint" id="analyticsHint" style="margin-top:8px; display:none"></div>
- <div class="hint" id="analyticsForgetRow" style="margin-top:8px; display:none">
- Anonymous install id: <code id="analyticsInstallId" style="font-size:11px"></code>
- <button type="button" class="ghost-btn" onclick="rotateAnalyticsId()"
- style="margin-left:8px; padding:2px 8px; font-size:11px">forget me</button>
- </div>
- </div>
-
<div class="settings-section" id="settingsCustomSection" style="display:none">
<h3>Custom (advanced)</h3>
<div class="settings-row" style="margin-bottom:10px">
@@ -25981,9 +25801,6 @@ async function openSettingsModal() {
// on the JS side; only ON/OFF gets persisted.
_spicyArmed = false;
renderSpicyState(!!cur.spicy_mode);
-
- // Anonymous analytics toggle — single click, no two-step confirm.
- renderAnalyticsState(!!cur.analytics_enabled, cur.analytics_install_id || '');
}
function renderMemoryPolicyHint() {
@@ -26085,103 +25902,6 @@ async function _persistSpicyMode(target) {
}
}
-// Anonymous analytics — single-click toggle (no two-click confirm; this
-// is opt-IN, not destructive). The install_id is generated server-side
-// the first time the flag flips ON; the UI just shows the bare boolean.
-function renderAnalyticsState(isOn, installId) {
- const badge = document.getElementById('analyticsStateBadge');
- const btn = document.getElementById('analyticsToggleBtn');
- const hint = document.getElementById('analyticsHint');
- const forgetRow = document.getElementById('analyticsForgetRow');
- const idEl = document.getElementById('analyticsInstallId');
- if (!badge || !btn) return;
- badge.classList.remove('on', 'armed');
- if (isOn) {
- badge.textContent = 'ON';
- badge.classList.add('on');
- btn.textContent = 'Disable';
- btn.classList.remove('primary-btn');
- btn.classList.add('ghost-btn');
- if (hint) {
- hint.style.display = '';
- hint.textContent = 'Thanks — anonymous render events are flowing. Toggle OFF to stop instantly.';
- }
- if (forgetRow && installId) {
- forgetRow.style.display = '';
- if (idEl) idEl.textContent = installId.slice(0, 8) + '…';
- } else if (forgetRow) {
- forgetRow.style.display = 'none';
- }
- } else {
- badge.textContent = 'OFF';
- btn.textContent = 'Enable anonymous analytics';
- btn.classList.remove('primary-btn');
- btn.classList.add('ghost-btn');
- if (hint) { hint.style.display = 'none'; hint.textContent = ''; }
- if (forgetRow) forgetRow.style.display = 'none';
- }
-}
-
-async function toggleAnalytics() {
- const cur = (_settingsCache && _settingsCache.settings) || {};
- const target = !cur.analytics_enabled;
- const status = document.getElementById('settingsStatus');
- try {
- const fd = new URLSearchParams();
- fd.set('analytics_enabled', target ? 'true' : 'false');
- const r = await fetch('/settings', { method: 'POST', body: fd });
- const j = await r.json();
- if (j.error) throw new Error(j.error);
- if (_settingsCache && _settingsCache.settings) {
- _settingsCache.settings.analytics_enabled = !!target;
- if (j.settings && j.settings.analytics_install_id) {
- _settingsCache.settings.analytics_install_id = j.settings.analytics_install_id;
- }
- }
- const id = (_settingsCache?.settings?.analytics_install_id) || '';
- renderAnalyticsState(!!target, id);
- if (status) {
- status.textContent = target ? 'Anonymous analytics ON · thanks for helping' : 'Anonymous analytics OFF';
- status.className = 'settings-status ok';
- }
- } catch (e) {
- if (status) {
- status.textContent = 'Could not change analytics: ' + (e.message || e);
- status.className = 'settings-status err';
- }
- }
-}
-
-async function rotateAnalyticsId() {
- // Clear the saved id; server regenerates a fresh one on next opt-in
- // (or on the next emit if still enabled). Sends one settings_opt_out
- // + one settings_opt_in with the new id, which is the right wire
- // shape — anyone correlating events sees a new identity from now on.
- const status = document.getElementById('settingsStatus');
- try {
- const fd = new URLSearchParams();
- fd.set('analytics_install_id', '');
- const r = await fetch('/settings', { method: 'POST', body: fd });
- const j = await r.json();
- if (j.error) throw new Error(j.error);
- if (_settingsCache && _settingsCache.settings) {
- _settingsCache.settings.analytics_install_id =
- (j.settings && j.settings.analytics_install_id) || '';
- }
- const id = _settingsCache?.settings?.analytics_install_id || '';
- renderAnalyticsState(!!(_settingsCache?.settings?.analytics_enabled), id);
- if (status) {
- status.textContent = 'Analytics identity reset — future events use a fresh anonymous id';
- status.className = 'settings-status ok';
- }
- } catch (e) {
- if (status) {
- status.textContent = 'Could not reset analytics id: ' + (e.message || e);
- status.className = 'settings-status err';
- }
- }
-}
-
function setTokenStatus(prefix, isSet, dirty) {
const el = document.getElementById(prefix + 'Status');
if (!el) return;
@@ -28681,31 +28401,6 @@ if __name__ == "__main__":
threading.Thread(target=version_check_loop, daemon=True).start()
else:
_detect_local_install_state()
- # ---- Anonymous telemetry — strictly opt-in (default OFF) ------------
- # Wire the analytics module to read settings live so the user can
- # flip the toggle without restarting. The install_id is generated
- # lazily the first time the user opts in (see /settings handler).
- try:
- import analytics as _ph_analytics # type: ignore
- _ph_analytics.install(
- is_enabled=lambda: bool(get_settings().get("analytics_enabled", False)),
- get_install_id=lambda: str(get_settings().get("analytics_install_id", "")),
- get_endpoint=lambda: (os.environ.get("PHOSPHENE_ANALYTICS_ENDPOINT", "").strip()
- or _ph_analytics.DEFAULT_ENDPOINT),
- log=push,
- )
- # Boot event. Carries anonymized fingerprint so we can correlate
- # bugs to hardware tier + OS without per-user identifiers.
- _ph_analytics.emit("panel_boot", {
- "version": _read_local_version() or "unknown",
- "sha": (_VERSION_STATE.get("local_short") or ""),
- "branch": (_VERSION_STATE.get("local_branch") or "?"),
- "tier": SYSTEM_CAPS.get("label", "?"),
- "allows_q8": bool(SYSTEM_CAPS.get("allows_q8", False)),
- **_ph_analytics.system_fingerprint(),
- })
- except Exception as _exc: # noqa: BLE001 — telemetry must never crash the panel
- sys.stderr.write(f"WARN: analytics setup failed silently: {_exc}\n")
# Pre-flight: bind in a try/except so a busy port surfaces an actionable
# one-liner instead of a 6-frame Python traceback. The bare OSError
# ("[Errno 48] Address already in use") was confusing users who'd closed
@@ -28728,11 +28423,3 @@ if __name__ == "__main__":
finally:
HELPER.kill()
caffeinate_off()
- # Telemetry is best-effort; daemon thread will die with the
- # process anyway, but a clean stop lets queued events flush
- # if the user closes the panel gracefully.
- try:
- import analytics as _ph_a # type: ignore
- _ph_a.shutdown()
- except Exception:
- pass