Staging (Week of 26-06-22)

anton/__init__.py

@@ -1 +1 @@
-__version__ = "2.26.6.22.1"
+__version__ = "2.26.6.30.1"

anton/analytics.py

@@ -35,6 +35,7 @@
 from __future__ import annotations
 
 import hashlib
+import os
 import threading
 import time
 import urllib.parse
@@ -51,6 +52,34 @@
 # a process, so computing it once is sufficient.
 _cached_aid: str | None = None
 
+# Cached CI detection. Env-derived (no PII), consistent with this module's
+# anonymous design. CI/automation traffic is dropped entirely (see send_event)
+# rather than tagged, so it can't pollute the product funnel. Driven by an
+# explicit Anton-owned signal (ANTON_IS_CI) with known provider markers as a
+# convenience fallback; the bare ``CI`` var is intentionally not consulted —
+# it's frequently set to "false" or leaks into local dev shells (ENG-385).
+_cached_is_ci: bool | None = None
+
+
+def _env_true(name: str) -> bool:
+    return os.environ.get(name, "").strip().lower() in {"1", "true", "yes", "on"}
+
+
+def _is_ci() -> bool:
+    """Return True for Anton automation/CI traffic (cached, env-only)."""
+    global _cached_is_ci
+    if _cached_is_ci is None:
+        _cached_is_ci = (
+            _env_true("ANTON_IS_CI")
+            or _env_true("GITHUB_ACTIONS")
+            or _env_true("GITLAB_CI")
+            or _env_true("BUILDKITE")
+            or _env_true("CIRCLECI")
+            or _env_true("TF_BUILD")
+            or bool(os.environ.get("JENKINS_URL"))
+        )
+    return _cached_is_ci
+
 
 def get_installation_id() -> str:
     """Return a deterministic, anonymous machine fingerprint.
@@ -97,10 +126,17 @@ def send_event(settings: "AntonSettings", action: str, **extra: str) -> None:
         settings: Resolved AntonSettings (checked for analytics_enabled / analytics_url).
         action: Event name, e.g. ``"anton_started"``.
         **extra: Additional key=value pairs appended as query parameters.
+
+    CI/automation traffic (ANTON_IS_CI, or a known CI provider) is dropped
+    rather than sent, so it can't pollute the product funnel (no PII either way).
     """
     try:
         if not settings.analytics_enabled:
             return
+        # Drop CI/automation traffic entirely — no value in product analytics
+        # from CI runs, and dropping avoids a per-query exclusion filter.
+        if _is_ci():
+            return
         url = settings.analytics_url
         if not url:
             return

anton/config/settings.py

@@ -8,14 +8,20 @@
 
 
 def _build_env_files() -> list[str]:
-    """Build .env loading chain: cwd/.env -> .anton/.env -> ~/.anton/.env"""
+    """Build .env loading chain: cwd/.env -> .anton/.env -> ~/.anton/.env
+    -> ~/.cowork/.env. Later files win, so the consolidated ~/.cowork/.env
+    takes precedence; ~/.anton/.env stays as a fallback for installs that
+    haven't migrated yet."""
     files: list[str] = [".env"]
     local_env = Path.cwd() / ".anton" / ".env"
     if local_env.is_file():
         files.append(str(local_env))
     user_env = Path("~/.anton/.env").expanduser()
     if user_env.is_file():
         files.append(str(user_env))
+    cowork_env = Path("~/.cowork/.env").expanduser()
+    if cowork_env.is_file():
+        files.append(str(cowork_env))
     return files
 
 
@@ -112,7 +118,7 @@ class AntonSettings(CoreSettings):
     minds_ssl_verify: bool = True
 
     # Publish service
-    publish_url: str = "https://4nton.ai"
+    publish_url: str = "https://view.mindshub.ai"
 
     backend: str = "local"  # local | remote
 
@@ -139,7 +145,18 @@ def model_post_init(self, __context) -> None:
         ):
             self.openai_api_key = self.minds_api_key
             if not self.openai_base_url:
-                self.openai_base_url = f"{self.minds_url.rstrip('/')}/api/v1"
+                # Host-aware base URL: api.mindshub.ai serves the
+                # OpenAI-compatible API at /v1, the legacy mdb.ai host at
+                # /api/v1. The previous hardcoded /api/v1 was correct only
+                # for mdb.ai and produced a wrong endpoint for mindshub
+                # (ENG-436). Mirrors cowork-server's minds_chat_base_url.
+                base = self.minds_url.rstrip("/")
+                if base.endswith("/v1"):
+                    self.openai_base_url = base
+                elif "mdb.ai" in base:
+                    self.openai_base_url = f"{base}/api/v1"
+                else:
+                    self.openai_base_url = f"{base}/v1"
 
     _workspace: Path = PrivateAttr(default=None)
 

anton/core/backends/local.py

@@ -352,7 +352,17 @@ async def start(self) -> None:
             and "ANTON_MINDS_URL" in env
             and self._coding_provider == "openai-compatible"
         ):
-            env["OPENAI_BASE_URL"] = f"{env['ANTON_MINDS_URL'].rstrip('/')}/api/v1"
+            # Host-aware (ENG-436): api.mindshub.ai serves /v1, legacy
+            # mdb.ai serves /api/v1. The previous hardcoded /api/v1 was
+            # wrong for mindshub. Mirrors config/settings.py +
+            # cowork-server minds_chat_base_url.
+            _minds_base = env["ANTON_MINDS_URL"].rstrip("/")
+            if _minds_base.endswith("/v1"):
+                env["OPENAI_BASE_URL"] = _minds_base
+            elif "mdb.ai" in _minds_base:
+                env["OPENAI_BASE_URL"] = f"{_minds_base}/api/v1"
+            else:
+                env["OPENAI_BASE_URL"] = f"{_minds_base}/v1"
         if self._coding_api_key:
             sdk_key = {
                 "anthropic": "ANTHROPIC_API_KEY",

anton/core/backends/scratchpad_boot.py

@@ -80,23 +80,26 @@ def _dump_namespace(ns: dict) -> str | None:
                 "ANTON_OPENAI_API_KEY"
             )
             _llm_api_version = os.environ.get("ANTON_OPENAI_API_VERSION") or None
-            # Mirror LLMClient.from_settings: openai-compatible endpoints
-            # (Minds/MindsHub) expect image blocks in Anthropic native format,
-            # not OpenAI image_url data-URLs.
             _llm_provider_kwargs: dict = {
                 "api_key": _llm_api_key or None,
                 "base_url": _llm_base_url or None,
                 "ssl_verify": _llm_ssl_verify,
                 "api_version": _llm_api_version,
             }
-            # Endpoints that proxy to Minds / MindsHub / MDB.AI
-            # speak Anthropic content format over the OpenAI HTTP envelope,
-            # so image blocks must be Anthropic-shaped, not OpenAI image_url.
+            # Only Minds / MindsHub / MDB.AI proxy Anthropic over the OpenAI HTTP
+            # envelope, so ONLY they need image blocks in Anthropic native format.
+            # Gate on the host, NOT on "openai-compatible" generally: other
+            # OpenAI-compatible endpoints (e.g. Gemini at
+            # generativelanguage.googleapis.com, or a generic proxy) expect
+            # standard OpenAI image_url blocks. Forcing anthropic format for all
+            # openai-compatible mangled images for Gemini. OpenAIProvider already
+            # defaults to supports_vision=True / vision_format="openai", so the
+            # non-proxy case needs no override.
             _llm_url_lower = (_llm_base_url or "").lower()
             _anthropic_proxy = any(
                 host in _llm_url_lower for host in ("mdb.ai", "mindshub.ai")
             )
-            if _scratchpad_provider_name == "openai-compatible" or _anthropic_proxy:
+            if _anthropic_proxy:
                 _llm_provider_kwargs["supports_vision"] = True
                 _llm_provider_kwargs["vision_format"] = "anthropic"
             # Resolve the OpenAI "flavor" so the injected web_search() helper can

anton/core/datasources/datasources.md

@@ -720,6 +720,111 @@ password if your provider requires it.
 
 ---
 
+## HasData
+
+Single API key for 25+ structured web data APIs: Google SERP/Maps/Trends/News/Shopping/Flights/Hotels,
+Amazon, Airbnb, Booking.com, Zillow, Redfin, YouTube, Instagram, Indeed, Glassdoor, Yelp,
+Yellow Pages, Shopify, Bing, and general web scraping.
+
+Base URL: `https://api.hasdata.com` · Auth: `x-api-key` header.
+
+**⚠️ Two-step async pattern — ALWAYS follow this:**
+1. `GET /scrape/<endpoint>?<params>` → response contains `requestMetadata.resultUrl` (a CDN URL)
+2. Wait 2–3 s, then `GET <resultUrl>` → actual JSON results
+
+Do not try to read results from the first response body — it only contains metadata.
+Do not hit the API root (`/`) — there is no handler there.
+
+```python
+# Canonical usage pattern
+import httpx, os, time
+
+key = os.environ.get('DS_API_KEY')
+base = os.environ.get('DS_BASE_URL', 'https://api.hasdata.com')
+headers = {"x-api-key": key}
+
+# Step 1 — trigger the scrape
+r = httpx.get(f"{base}/scrape/google/serp", headers=headers, params={"q": "your query", "gl": "us"})
+r.raise_for_status()
+result_url = r.json()["requestMetadata"]["resultUrl"]
+
+# Step 2 — wait for CDN, then fetch results
+time.sleep(3)
+data = httpx.get(result_url).json()
+
+# Guard sitelinks — can be a dict OR a list depending on the result
+sitelinks = data.get("sitelinks", [])
+if isinstance(sitelinks, list):
+    sitelinks = sitelinks[:3]
+```
+
+```yaml
+engine: hasdata
+display_name: HasData
+name_from: api_key
+popular: true
+fields:
+  - { name: api_key,  required: true,  secret: true,  description: "HasData API key from hasdata.com dashboard" }
+  - { name: base_url, required: false, secret: false, description: "API base URL", default: "https://api.hasdata.com" }
+test_snippet: |
+  import httpx, os, time
+  key = os.environ['DS_API_KEY']
+  base = os.environ.get('DS_BASE_URL', 'https://api.hasdata.com')
+  headers = {"x-api-key": key}
+  r = httpx.get(f"{base}/scrape/google/serp", headers=headers, params={"q": "test", "num": "1"})
+  assert r.status_code == 200, f"unexpected status {r.status_code}: {r.text[:200]}"
+  meta = r.json().get("requestMetadata", {})
+  assert "resultUrl" in meta, f"missing resultUrl in response: {list(meta.keys())}"
+  time.sleep(3)
+  results = httpx.get(meta["resultUrl"]).json()
+  assert isinstance(results, dict), f"unexpected results type: {type(results)}"
+  print("ok")
+```
+
+Endpoints (all GET, auth via `x-api-key` header):
+
+| API | Path | Key params |
+|---|---|---|
+| Google SERP | `/scrape/google/serp` | `q`, `gl`, `hl`, `location`, `num`, `start`, `deviceType` |
+| Google AI Mode | `/scrape/google/ai-overview` | `pageToken` (from SERP aiOverview block) |
+| Google Maps Search | `/scrape/google-maps/search` | `q`, `gl`, `hl`, `ll`, `start` |
+| Google Maps Reviews | `/scrape/google-maps/reviews` | `placeId`, `hl` |
+| Google Maps Photos | `/scrape/google-maps/photos` | `placeId` |
+| Google Maps Posts | `/scrape/google-maps/posts` | `placeId` |
+| Google Trends | `/scrape/google/trends` | `q`, `geo`, `date` |
+| Google News | `/scrape/google/news` | `q`, `gl`, `hl` |
+| Google Shopping | `/scrape/google/shopping` | `q`, `gl`, `hl` |
+| Google Images | `/scrape/google/images` | `q`, `gl`, `hl` |
+| Google Flights | `/scrape/google/flights` | `departureId`, `arrivalId`, `outboundDate`, `returnDate`, `type` |
+| Google Hotels | `/scrape/google/hotels` | `q`, `checkIn`, `checkOut`, `adults`, `gl`, `hl` |
+| Google Rank Checker | `/scrape/google/rank-checker` | `q`, `website`, `gl`, `hl` |
+| Bing Search | `/scrape/bing/serp` | `q`, `count`, `offset`, `mkt` |
+| Amazon Product | `/scrape/amazon/product` | `asin`, `domain`, `deliveryZip` |
+| Amazon Search | `/scrape/amazon/search` | `q`, `domain`, `page` |
+| Airbnb Listings | `/scrape/airbnb/search` | `location`, `checkIn`, `checkOut`, `adults` |
+| Airbnb Property | `/scrape/airbnb/listing` | `listingId`, `checkIn`, `checkOut` |
+| Booking.com Search | `/scrape/booking/search` | `location`, `checkIn`, `checkOut`, `adults` |
+| Google Hotels (Booking) | `/scrape/booking/place` | `placeId` |
+| Glassdoor Jobs | `/scrape/glassdoor/jobs` | `q`, `location` |
+| Indeed Jobs | `/scrape/indeed/jobs` | `q`, `location`, `start` |
+| Instagram Profile | `/scrape/instagram/profile` | `username` |
+| YouTube Search | `/scrape/youtube/search` | `q` |
+| YouTube Video | `/scrape/youtube/video` | `videoId` |
+| YouTube Channel | `/scrape/youtube/channel` | `channelId` |
+| YouTube Transcript | `/scrape/youtube/transcript` | `videoId`, `lang` |
+| Redfin Listings | `/scrape/redfin/search` | `location` |
+| Redfin Property | `/scrape/redfin/property` | `url` |
+| Zillow Listings | `/scrape/zillow/search` | `location` |
+| Zillow Property | `/scrape/zillow/property` | `zpid` |
+| Shopify Products | `/scrape/shopify/products` | `domain` |
+| Yelp Search | `/scrape/yelp/search` | `q`, `location` |
+| Yelp Place | `/scrape/yelp/place` | `alias` |
+| Yellow Pages Search | `/scrape/yellowpages/search` | `q`, `location` |
+| Yellow Pages Place | `/scrape/yellowpages/place` | `url` |
+| Web Scraper | `/scrape/web` | `url`, `extractRules`, `screenshot` |
+
+---
+
 ## Adding a new data source
 
 Follow the YAML format above. Add to `~/.anton/datasources.md` (user overrides).

anton/core/interaction/__init__.py

@@ -0,0 +1,15 @@
+"""Mid-turn human interaction primitives for the agent loop.
+
+Currently this package hosts file/folder disambiguation (the ``select_path``
+tool). The agent depends only on the abstract :class:`SelectionElicitor`
+strategy; each host (CLI, cowork-server harness, …) supplies a concrete
+implementation, so the core never learns how the prompt is surfaced.
+"""
+
+from anton.core.interaction.selection import (
+    SelectionElicitor,
+    SelectionOption,
+    SelectionRequest,
+)
+
+__all__ = ["SelectionElicitor", "SelectionOption", "SelectionRequest"]

anton/core/interaction/cli.py

@@ -0,0 +1,53 @@
+"""Terminal implementation of :class:`SelectionElicitor` for standalone CLI runs.
+
+Renders a numbered picker to the console and reads the user's choice. The agent
+loop already pauses the escape watcher around interactive tools, so this only
+has to print the options and await a number.
+"""
+
+from __future__ import annotations
+
+from anton.core.interaction.selection import SelectionRequest
+
+__all__ = ["CLISelectionElicitor"]
+
+
+class CLISelectionElicitor:
+    """Picker that prompts on the terminal (standalone ``anton`` chat)."""
+
+    def __init__(self, console) -> None:
+        self._console = console
+
+    async def elicit(self, request: SelectionRequest) -> str | None:
+        from anton.utils.prompt import prompt_or_cancel
+
+        # Browse mode has no visual file tree on a terminal — fall back to a
+        # typed path (the GUI host gets a real navigable browser instead).
+        if request.mode == "browse":
+            self._console.print(f"\n[bold]{request.prompt}[/]")
+            if request.root:
+                self._console.print(f"  [dim]starting at {request.root}[/]")
+            chosen = await prompt_or_cancel("Enter a path (Esc to cancel)")
+            return (chosen or "").strip() or None
+
+        options = request.options
+        if not options:
+            return None
+
+        self._console.print(f"\n[bold]{request.prompt}[/]")
+        for index, option in enumerate(options, start=1):
+            icon = "📁" if option.kind == "folder" else "📄"
+            detail = f"  [dim]{option.detail}[/]" if option.detail else ""
+            self._console.print(f"  [bold]{index}[/]. {icon} {option.label}{detail}")
+
+        choice = await prompt_or_cancel(
+            "Select a number (Esc to cancel)",
+            choices=[str(i) for i in range(1, len(options) + 1)],
+        )
+        if choice is None:
+            return None
+        try:
+            selected = int(choice) - 1
+        except ValueError:
+            return None
+        return options[selected].value if 0 <= selected < len(options) else None