compair-core 0.3.15__tar.gz → 0.4.1__tar.gz

{compair_core-0.3.15 → compair_core-0.4.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compair-core
-Version: 0.3.15
+Version: 0.4.1
 Summary: Open-source foundation of the Compair collaboration platform.
 Author: RocketResearch, Inc.
 License: MIT
@@ -86,7 +86,8 @@ Container definitions and build pipelines live outside this public package:
 Key environment variables for the core edition:
 
 - `COMPAIR_EDITION` (`core`) – corresponds to this core local implementation.
-- `COMPAIR_SQLITE_DIR` / `COMPAIR_SQLITE_NAME` – override the default local SQLite path (falls back to `./compair_data` if `/data` is not writable).
+- `COMPAIR_DATABASE_URL` – optional explicit SQLAlchemy URL (e.g. `postgresql+psycopg2://user:pass@host/db`). When omitted, Compair falls back to a local SQLite file.
+- `COMPAIR_DB_DIR` / `COMPAIR_DB_NAME` – directory and filename for the bundled SQLite database (default: `~/.compair-core/data/compair.db`). Legacy `COMPAIR_SQLITE_*` variables remain supported.
 - `COMPAIR_LOCAL_MODEL_URL` – endpoint for your local embeddings/feedback service (defaults to `http://local-model:9000`).
 - `COMPAIR_EMAIL_BACKEND` – the core mailer logs emails to stdout; cloud overrides this with transactional delivery.
 - `COMPAIR_REQUIRE_AUTHENTICATION` (`true`) – set to `false` to run the API in single-user mode without login or account management. When disabled, Compair auto-provisions a local user, group, and long-lived session token so you can upload documents immediately.
@@ -94,8 +95,10 @@ Key environment variables for the core edition:
 - `COMPAIR_INCLUDE_LEGACY_ROUTES` (`false`) – opt-in to the full legacy API surface (used by the hosted product) when running the core edition. Leave unset to expose only the streamlined single-user endpoints in Swagger.
 - `COMPAIR_EMBEDDING_DIM` – force the embedding vector size stored in the database (defaults to 384 for core, 1536 for cloud). Keep this in sync with whichever embedding model you configure.
 - `COMPAIR_VECTOR_BACKEND` (`auto`) – set to `pgvector` when running against PostgreSQL with the pgvector extension, or `json` to store embeddings as JSON (the default for SQLite deployments).
-- `COMPAIR_GENERATION_PROVIDER` (`local`) – choose how feedback is produced. Options: `local` (call the bundled FastAPI service), `openai` (use ChatGPT-compatible APIs with an API key), or `fallback` (skip generation and surface similar references only).
+- `COMPAIR_GENERATION_PROVIDER` (`local`) – choose how feedback is produced. Options: `local` (call the bundled FastAPI service), `openai` (use ChatGPT-compatible APIs with an API key), `http` (POST the request to a custom endpoint), or `fallback` (skip generation and surface similar references only).
 - `COMPAIR_OPENAI_API_KEY` / `COMPAIR_OPENAI_MODEL` – when using the OpenAI provider, supply your API key and optional model name (defaults to `gpt-4o-mini`). The fallback kicks in automatically if the key or SDK is unavailable.
+- `COMPAIR_GENERATION_ENDPOINT` – HTTP endpoint invoked when `COMPAIR_GENERATION_PROVIDER=http`; the service receives a JSON payload (`document`, `references`, `length_instruction`) and should return `{"feedback": ...}`.
+- `COMPAIR_OCR_ENDPOINT` – endpoint the backend calls for OCR uploads (defaults to the bundled Tesseract wrapper at `http://local-ocr:9001/ocr-file`). Provide your own service by overriding this URL.
 
 See `compair_core/server/settings.py` for the full settings surface.
 

{compair_core-0.3.15 → compair_core-0.4.1}/README.md

@@ -51,7 +51,8 @@ Container definitions and build pipelines live outside this public package:
 Key environment variables for the core edition:
 
 - `COMPAIR_EDITION` (`core`) – corresponds to this core local implementation.
-- `COMPAIR_SQLITE_DIR` / `COMPAIR_SQLITE_NAME` – override the default local SQLite path (falls back to `./compair_data` if `/data` is not writable).
+- `COMPAIR_DATABASE_URL` – optional explicit SQLAlchemy URL (e.g. `postgresql+psycopg2://user:pass@host/db`). When omitted, Compair falls back to a local SQLite file.
+- `COMPAIR_DB_DIR` / `COMPAIR_DB_NAME` – directory and filename for the bundled SQLite database (default: `~/.compair-core/data/compair.db`). Legacy `COMPAIR_SQLITE_*` variables remain supported.
 - `COMPAIR_LOCAL_MODEL_URL` – endpoint for your local embeddings/feedback service (defaults to `http://local-model:9000`).
 - `COMPAIR_EMAIL_BACKEND` – the core mailer logs emails to stdout; cloud overrides this with transactional delivery.
 - `COMPAIR_REQUIRE_AUTHENTICATION` (`true`) – set to `false` to run the API in single-user mode without login or account management. When disabled, Compair auto-provisions a local user, group, and long-lived session token so you can upload documents immediately.
@@ -59,8 +60,10 @@ Key environment variables for the core edition:
 - `COMPAIR_INCLUDE_LEGACY_ROUTES` (`false`) – opt-in to the full legacy API surface (used by the hosted product) when running the core edition. Leave unset to expose only the streamlined single-user endpoints in Swagger.
 - `COMPAIR_EMBEDDING_DIM` – force the embedding vector size stored in the database (defaults to 384 for core, 1536 for cloud). Keep this in sync with whichever embedding model you configure.
 - `COMPAIR_VECTOR_BACKEND` (`auto`) – set to `pgvector` when running against PostgreSQL with the pgvector extension, or `json` to store embeddings as JSON (the default for SQLite deployments).
-- `COMPAIR_GENERATION_PROVIDER` (`local`) – choose how feedback is produced. Options: `local` (call the bundled FastAPI service), `openai` (use ChatGPT-compatible APIs with an API key), or `fallback` (skip generation and surface similar references only).
+- `COMPAIR_GENERATION_PROVIDER` (`local`) – choose how feedback is produced. Options: `local` (call the bundled FastAPI service), `openai` (use ChatGPT-compatible APIs with an API key), `http` (POST the request to a custom endpoint), or `fallback` (skip generation and surface similar references only).
 - `COMPAIR_OPENAI_API_KEY` / `COMPAIR_OPENAI_MODEL` – when using the OpenAI provider, supply your API key and optional model name (defaults to `gpt-4o-mini`). The fallback kicks in automatically if the key or SDK is unavailable.
+- `COMPAIR_GENERATION_ENDPOINT` – HTTP endpoint invoked when `COMPAIR_GENERATION_PROVIDER=http`; the service receives a JSON payload (`document`, `references`, `length_instruction`) and should return `{"feedback": ...}`.
+- `COMPAIR_OCR_ENDPOINT` – endpoint the backend calls for OCR uploads (defaults to the bundled Tesseract wrapper at `http://local-ocr:9001/ocr-file`). Provide your own service by overriding this URL.
 
 See `compair_core/server/settings.py` for the full settings surface.
 

{compair_core-0.3.15 → compair_core-0.4.1}/compair_core/api.py

@@ -2370,6 +2370,8 @@ def get_activity_feed(
 ):
     """Retrieve recent activities for a user's groups."""
     require_feature(HAS_ACTIVITY, "Activity feed")
+    if not IS_CLOUD:
+        raise HTTPException(status_code=501, detail="Activity feed is only available in the Compair Cloud edition.")
     with compair.Session() as session:
         # Get user's groups
         
@@ -3514,7 +3516,11 @@ CORE_PATHS: set[str] = {
     "/load_documents",
     "/load_document",
     "/load_document_by_id",
+    "/load_user_files",
     "/create_doc",
+    "/update_doc",
+    "/delete_doc",
+    "/delete_docs",
     "/process_doc",
     "/status/{task_id}",
     "/upload/ocr-file",
@@ -3523,6 +3529,7 @@ CORE_PATHS: set[str] = {
     "/load_references",
     "/load_feedback",
     "/documents/{document_id}/feedback",
+    "/get_activity_feed",
 }
 
 for route in router.routes:

{compair_core-0.3.15 → compair_core-0.4.1}/compair_core/compair/__init__.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import os
+from pathlib import Path
 from sqlalchemy import Engine, create_engine
 from sqlalchemy.orm import sessionmaker
 
@@ -37,27 +38,51 @@ if edition == "cloud":
 
 
 def _handle_engine() -> Engine:
+    # Preferred configuration: explicit database URL
+    explicit_url = (
+        os.getenv("COMPAIR_DATABASE_URL")
+        or os.getenv("COMPAIR_DB_URL")
+        or os.getenv("DATABASE_URL")
+    )
+    if explicit_url:
+        if explicit_url.startswith("sqlite:"):
+            return create_engine(explicit_url, connect_args={"check_same_thread": False})
+        return create_engine(explicit_url)
+
+    # Backwards compatibility with legacy Postgres env variables
     db = os.getenv("DB")
     db_user = os.getenv("DB_USER")
     db_passw = os.getenv("DB_PASSW")
-    db_url = os.getenv("DB_URL")
+    db_host = os.getenv("DB_URL")
 
-    if all([db, db_user, db_passw, db_url]):
+    if all([db, db_user, db_passw, db_host]):
         return create_engine(
-            f"postgresql+psycopg2://{db_user}:{db_passw}@{db_url}/{db}",
+            f"postgresql+psycopg2://{db_user}:{db_passw}@{db_host}/{db}",
             pool_size=10,
             max_overflow=0,
         )
 
-    sqlite_dir = os.getenv("COMPAIR_SQLITE_DIR", "/data")
+    # Local default: place an SQLite database inside COMPAIR_DB_DIR
+    db_dir = (
+        os.getenv("COMPAIR_DB_DIR")
+        or os.getenv("COMPAIR_SQLITE_DIR")
+        or os.path.join(Path.home(), ".compair-core", "data")
+    )
+    db_name = os.getenv("COMPAIR_DB_NAME") or os.getenv("COMPAIR_SQLITE_NAME") or "compair.db"
+
+    db_path = Path(db_dir).expanduser()
     try:
-        os.makedirs(sqlite_dir, exist_ok=True)
+        db_path.mkdir(parents=True, exist_ok=True)
     except OSError:
-        fallback_dir = os.path.join(os.getcwd(), "compair_data")
-        os.makedirs(fallback_dir, exist_ok=True)
-        sqlite_dir = fallback_dir
-    sqlite_path = os.path.join(sqlite_dir, os.getenv("COMPAIR_SQLITE_NAME", "compair.db"))
-    return create_engine(f"sqlite:///{sqlite_path}", connect_args={"check_same_thread": False})
+        fallback_dir = Path(os.getcwd()) / "compair_data"
+        fallback_dir.mkdir(parents=True, exist_ok=True)
+        db_path = fallback_dir
+
+    sqlite_path = db_path / db_name
+    return create_engine(
+        f"sqlite:///{sqlite_path}",
+        connect_args={"check_same_thread": False},
+    )
 
 
 def initialize_database() -> None:

compair_core-0.4.1/compair_core/compair/feedback.py

@@ -0,0 +1,313 @@
+from __future__ import annotations
+
+import os
+from typing import Any, Iterable, List
+
+import requests
+
+from .logger import log_event
+from .models import Document, User
+
+try:
+    import openai  # type: ignore
+except ImportError:  # pragma: no cover - optional dependency
+    openai = None  # type: ignore
+
+try:
+    from compair_cloud.feedback import Reviewer as CloudReviewer  # type: ignore
+    from compair_cloud.feedback import get_feedback as cloud_get_feedback  # type: ignore
+except (ImportError, ModuleNotFoundError):
+    CloudReviewer = None  # type: ignore
+    cloud_get_feedback = None  # type: ignore
+
+
+class Reviewer:
+    """Edition-aware wrapper that selects a feedback provider based on configuration."""
+
+    def __init__(self) -> None:
+        self.edition = os.getenv("COMPAIR_EDITION", "core").lower()
+        self.provider = os.getenv("COMPAIR_GENERATION_PROVIDER", "local").lower()
+        self.length_map = {
+            "Brief": "1–2 short sentences",
+            "Detailed": "A couple short paragraphs",
+            "Verbose": "As thorough as reasonably possible without repeating information",
+        }
+
+        self._cloud_impl = None
+        self._openai_client = None
+        self.openai_model = os.getenv("COMPAIR_OPENAI_MODEL", "gpt-5-nano")
+        self.custom_endpoint = os.getenv("COMPAIR_GENERATION_ENDPOINT")
+
+        if self.edition == "cloud" and CloudReviewer is not None:
+            self._cloud_impl = CloudReviewer()
+            self.provider = "cloud"
+        else:
+            if self.provider == "openai":
+                api_key = os.getenv("COMPAIR_OPENAI_API_KEY")
+                if api_key and openai is not None:
+                    # Support both legacy (ChatCompletion) and new SDKs
+                    if hasattr(openai, "api_key"):
+                        openai.api_key = api_key  # type: ignore[assignment]
+                    if hasattr(openai, "OpenAI"):
+                        try:  # pragma: no cover - optional runtime dependency
+                            self._openai_client = openai.OpenAI(api_key=api_key)  # type: ignore[attr-defined]
+                        except Exception:  # pragma: no cover - if instantiation fails
+                            self._openai_client = None
+                if self._openai_client is None and not hasattr(openai, "ChatCompletion"):
+                    log_event("openai_feedback_unavailable", reason="openai_library_missing")
+                    self.provider = "fallback"
+            if self.provider == "http" and not self.custom_endpoint:
+                log_event("custom_feedback_unavailable", reason="missing_endpoint")
+                self.provider = "fallback"
+            if self.provider == "local":
+                self.model = os.getenv("COMPAIR_LOCAL_GENERATION_MODEL", "local-feedback")
+                base_url = os.getenv("COMPAIR_LOCAL_MODEL_URL", "http://local-model:9000")
+                route = os.getenv("COMPAIR_LOCAL_GENERATION_ROUTE", "/generate")
+                self.endpoint = f"{base_url.rstrip('/')}{route}"
+            else:
+                self.model = "external"
+                self.endpoint = None
+            if self.provider not in {"local", "openai", "http", "fallback"}:
+                log_event("feedback_provider_unknown", provider=self.provider)
+                self.provider = "fallback"
+
+    @property
+    def is_cloud(self) -> bool:
+        return self._cloud_impl is not None
+
+
+def _reference_snippets(references: Iterable[Any], limit: int = 3) -> List[str]:
+    snippets: List[str] = []
+    for ref in references:
+        snippet = getattr(ref, "content", "") or ""
+        snippet = snippet.replace("\n", " ").strip()
+        if snippet:
+            snippets.append(snippet[:200])
+        if len(snippets) == limit:
+            break
+    return snippets
+
+
+def _fallback_feedback(text: str, references: list[Any]) -> str:
+    snippets = _reference_snippets(references)
+    if not snippets:
+        return "NONE"
+    joined = "; ".join(snippets)
+    return f"Consider aligning with these reference passages: {joined}"
+
+
+
+def _local_reference_feedback(
+    reviewer: Reviewer,
+    references: list[Any],
+    user: User,
+) -> str | None:
+    if not references:
+        return None
+    summaries: list[str] = []
+    for ref in references[:3]:
+        doc = getattr(ref, "document", None)
+        title = getattr(doc, "title", None) or "a related document"
+        snippet = getattr(ref, "content", "") or getattr(ref, "text", "")
+        snippet = snippet.replace("\n", " ").strip()
+        if not snippet:
+            continue
+        summaries.append(f'"{title}" — {snippet[:200]}')
+    if not summaries:
+        return None
+    instruction = reviewer.length_map.get(user.preferred_feedback_length, "1–2 short sentences")
+    if len(summaries) == 1:
+        body = summaries[0]
+    else:
+        body = "; ".join(summaries)
+    return f"[local-feedback] {instruction}: Consider the guidance from {body}"
+
+
+def _openai_feedback(
+    reviewer: Reviewer,
+    doc: Document,
+    text: str,
+    references: list[Any],
+    user: User,
+) -> str | None:
+    if openai is None:
+        return None
+    instruction = reviewer.length_map.get(user.preferred_feedback_length, "1–2 short sentences")
+    ref_text = "\n\n".join(_reference_snippets(references, limit=3))
+    system_prompt = (
+        "You are Compair, an assistant that delivers concise, actionable feedback on a user's document. "
+        "Focus on clarity, cohesion, and usefulness."
+    )
+    user_prompt = (
+        f"Document:\n{text}\n\nHelpful reference excerpts:\n{ref_text or 'None provided'}\n\n"
+        f"Respond with {instruction} that highlights the most valuable revision to make next."
+    )
+
+    def _extract_response_text(response: Any) -> str | None:
+        if response is None:
+            return None
+        text_out = getattr(response, "output_text", None)
+        if isinstance(text_out, str) and text_out.strip():
+            return text_out.strip()
+        outputs = getattr(response, "output", None) or getattr(response, "outputs", None)
+        pieces: list[str] = []
+        if outputs:
+            for item in outputs:
+                content_field = None
+                if isinstance(item, dict):
+                    content_field = item.get("content")
+                else:
+                    content_field = getattr(item, "content", None)
+                if not content_field:
+                    continue
+                for part in content_field:
+                    if isinstance(part, dict):
+                        val = part.get("text") or part.get("output_text")
+                        if val:
+                            pieces.append(str(val))
+                    elif part:
+                        pieces.append(str(part))
+        if pieces:
+            merged = "\n".join(pieces).strip()
+            return merged or None
+        return None
+
+    try:
+        client = reviewer._openai_client
+        if client is None and hasattr(openai, "OpenAI"):
+            api_key = os.getenv("COMPAIR_OPENAI_API_KEY") or None
+            try:  # pragma: no cover - optional dependency differences
+                client = openai.OpenAI(api_key=api_key) if api_key else openai.OpenAI()
+            except TypeError:
+                client = openai.OpenAI()
+            reviewer._openai_client = client
+
+        content: str | None = None
+        if client is not None and hasattr(client, "responses"):
+            response = client.responses.create(
+                model=reviewer.openai_model,
+                instructions=system_prompt,
+                input=user_prompt,
+                max_output_tokens=256,
+                store=False,
+            )
+            content = _extract_response_text(response)
+        elif client is not None and hasattr(client, "chat") and hasattr(client.chat, "completions"):
+            response = client.chat.completions.create(
+                model=reviewer.openai_model,
+                messages=[{"role": "system", "content": system_prompt}, {"role": "user", "content": user_prompt}],
+                temperature=0.3,
+                max_tokens=256,
+            )
+            choices = getattr(response, "choices", None) or []
+            if choices:
+                message = getattr(choices[0], "message", None)
+                if message is not None:
+                    content = getattr(message, "content", None)
+                if not content:
+                    content = getattr(choices[0], "text", None)
+                if isinstance(content, str):
+                    content = content.strip()
+        elif hasattr(openai, "ChatCompletion"):
+            chat_response = openai.ChatCompletion.create(  # type: ignore[attr-defined]
+                model=reviewer.openai_model,
+                messages=[{"role": "system", "content": system_prompt}, {"role": "user", "content": user_prompt}],
+                temperature=0.3,
+                max_tokens=256,
+            )
+            content = chat_response["choices"][0]["message"]["content"].strip()  # type: ignore[index, assignment]
+        if content:
+            return content.strip()
+    except Exception as exc:  # pragma: no cover - network/API failure
+        log_event("openai_feedback_failed", error=str(exc))
+    return None
+
+
+def _local_feedback(
+    reviewer: Reviewer,
+    text: str,
+    references: list[Any],
+    user: User,
+) -> str | None:
+    payload = {
+        "document": text,
+        "references": [getattr(ref, "content", "") for ref in references],
+        "length_instruction": reviewer.length_map.get(
+            user.preferred_feedback_length,
+            "1–2 short sentences",
+        ),
+    }
+
+    try:
+        response = requests.post(reviewer.endpoint, json=payload, timeout=30)
+        response.raise_for_status()
+        data = response.json()
+        feedback = data.get("feedback") or data.get("text")
+        if feedback:
+            return str(feedback).strip()
+    except Exception as exc:  # pragma: no cover - network failures stay graceful
+        log_event("local_feedback_failed", error=str(exc))
+
+    return None
+
+
+def _http_feedback(
+    reviewer: Reviewer,
+    text: str,
+    references: list[Any],
+    user: User,
+) -> str | None:
+    if not reviewer.custom_endpoint:
+        return None
+    payload = {
+        "document": text,
+        "references": [getattr(ref, "content", "") for ref in references],
+        "length_instruction": reviewer.length_map.get(
+            user.preferred_feedback_length,
+            "1–2 short sentences",
+        ),
+    }
+    try:
+        response = requests.post(reviewer.custom_endpoint, json=payload, timeout=30)
+        response.raise_for_status()
+        data = response.json()
+        feedback = data.get("feedback") or data.get("text")
+        if isinstance(feedback, str):
+            feedback = feedback.strip()
+        if feedback:
+            return feedback
+    except Exception as exc:  # pragma: no cover - network failures stay graceful
+        log_event("custom_feedback_failed", error=str(exc))
+    return None
+
+
+def get_feedback(
+    reviewer: Reviewer,
+    doc: Document,
+    text: str,
+    references: list[Any],
+    user: User,
+) -> str:
+    if reviewer.is_cloud and cloud_get_feedback is not None:
+        return cloud_get_feedback(reviewer._cloud_impl, doc, text, references, user)  # type: ignore[arg-type]
+
+    if reviewer.provider == "openai":
+        feedback = _openai_feedback(reviewer, doc, text, references, user)
+        if feedback:
+            return feedback
+
+    if reviewer.provider == "http":
+        feedback = _http_feedback(reviewer, text, references, user)
+        if feedback:
+            return feedback
+
+    if reviewer.provider == "local":
+        feedback = _local_reference_feedback(reviewer, references, user)
+        if feedback:
+            return feedback
+        if getattr(reviewer, "endpoint", None):
+            feedback = _local_feedback(reviewer, text, references, user)
+            if feedback:
+                return feedback
+
+    return _fallback_feedback(text, references)

compair_core-0.4.1/compair_core/server/local_model/ocr.py

@@ -0,0 +1,44 @@
+"""Minimal OCR endpoint leveraging pytesseract when available."""
+from __future__ import annotations
+
+import io
+import os
+from typing import Any, Dict
+
+from fastapi import FastAPI, File, HTTPException, UploadFile
+
+app = FastAPI(title="Compair Local OCR", version="0.1.0")
+
+try:  # Optional dependency
+    import pytesseract  # type: ignore
+    from PIL import Image  # type: ignore
+except ImportError:  # pragma: no cover - optional
+    pytesseract = None  # type: ignore
+    Image = None  # type: ignore
+
+_OCR_FALLBACK = os.getenv("COMPAIR_LOCAL_OCR_FALLBACK", "text")  # text | none
+
+
+def _extract_text(data: bytes) -> str:
+    if pytesseract is None or Image is None:
+        if _OCR_FALLBACK == "text":
+            try:
+                return data.decode("utf-8")
+            except UnicodeDecodeError:
+                return data.decode("latin-1", errors="ignore")
+        return ""
+    try:
+        image = Image.open(io.BytesIO(data))
+        return pytesseract.image_to_string(image)
+    except Exception:
+        return ""
+
+
+@app.post("/ocr-file")
+async def ocr_file(file: UploadFile = File(...)) -> Dict[str, Any]:
+    payload = await file.read()
+    text = _extract_text(payload)
+    if not text:
+        raise HTTPException(status_code=501, detail="OCR not available or failed to extract text.")
+    return {"extracted_text": text}
+

{compair_core-0.3.15 → compair_core-0.4.1}/compair_core/server/routers/capabilities.py

@@ -36,6 +36,10 @@ def capabilities(settings: Settings = Depends(get_settings)) -> dict[str, object
             "docs": None if edition == "core" else 100,
             "feedback_per_day": None if edition == "core" else 50,
         },
+        "features": {
+            "ocr_upload": settings.ocr_enabled,
+            "activity_feed": edition == "cloud",
+        },
         "server": "Compair Cloud" if edition == "cloud" else "Compair Core",
         "version": settings.version,
         "legacy_routes": settings.include_legacy_routes,

{compair_core-0.3.15 → compair_core-0.4.1}/compair_core.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compair-core
-Version: 0.3.15
+Version: 0.4.1
 Summary: Open-source foundation of the Compair collaboration platform.
 Author: RocketResearch, Inc.
 License: MIT
@@ -86,7 +86,8 @@ Container definitions and build pipelines live outside this public package:
 Key environment variables for the core edition:
 
 - `COMPAIR_EDITION` (`core`) – corresponds to this core local implementation.
-- `COMPAIR_SQLITE_DIR` / `COMPAIR_SQLITE_NAME` – override the default local SQLite path (falls back to `./compair_data` if `/data` is not writable).
+- `COMPAIR_DATABASE_URL` – optional explicit SQLAlchemy URL (e.g. `postgresql+psycopg2://user:pass@host/db`). When omitted, Compair falls back to a local SQLite file.
+- `COMPAIR_DB_DIR` / `COMPAIR_DB_NAME` – directory and filename for the bundled SQLite database (default: `~/.compair-core/data/compair.db`). Legacy `COMPAIR_SQLITE_*` variables remain supported.
 - `COMPAIR_LOCAL_MODEL_URL` – endpoint for your local embeddings/feedback service (defaults to `http://local-model:9000`).
 - `COMPAIR_EMAIL_BACKEND` – the core mailer logs emails to stdout; cloud overrides this with transactional delivery.
 - `COMPAIR_REQUIRE_AUTHENTICATION` (`true`) – set to `false` to run the API in single-user mode without login or account management. When disabled, Compair auto-provisions a local user, group, and long-lived session token so you can upload documents immediately.
@@ -94,8 +95,10 @@ Key environment variables for the core edition:
 - `COMPAIR_INCLUDE_LEGACY_ROUTES` (`false`) – opt-in to the full legacy API surface (used by the hosted product) when running the core edition. Leave unset to expose only the streamlined single-user endpoints in Swagger.
 - `COMPAIR_EMBEDDING_DIM` – force the embedding vector size stored in the database (defaults to 384 for core, 1536 for cloud). Keep this in sync with whichever embedding model you configure.
 - `COMPAIR_VECTOR_BACKEND` (`auto`) – set to `pgvector` when running against PostgreSQL with the pgvector extension, or `json` to store embeddings as JSON (the default for SQLite deployments).
-- `COMPAIR_GENERATION_PROVIDER` (`local`) – choose how feedback is produced. Options: `local` (call the bundled FastAPI service), `openai` (use ChatGPT-compatible APIs with an API key), or `fallback` (skip generation and surface similar references only).
+- `COMPAIR_GENERATION_PROVIDER` (`local`) – choose how feedback is produced. Options: `local` (call the bundled FastAPI service), `openai` (use ChatGPT-compatible APIs with an API key), `http` (POST the request to a custom endpoint), or `fallback` (skip generation and surface similar references only).
 - `COMPAIR_OPENAI_API_KEY` / `COMPAIR_OPENAI_MODEL` – when using the OpenAI provider, supply your API key and optional model name (defaults to `gpt-4o-mini`). The fallback kicks in automatically if the key or SDK is unavailable.
+- `COMPAIR_GENERATION_ENDPOINT` – HTTP endpoint invoked when `COMPAIR_GENERATION_PROVIDER=http`; the service receives a JSON payload (`document`, `references`, `length_instruction`) and should return `{"feedback": ...}`.
+- `COMPAIR_OCR_ENDPOINT` – endpoint the backend calls for OCR uploads (defaults to the bundled Tesseract wrapper at `http://local-ocr:9001/ocr-file`). Provide your own service by overriding this URL.
 
 See `compair_core/server/settings.py` for the full settings surface.
 

{compair_core-0.3.15 → compair_core-0.4.1}/compair_core.egg-info/SOURCES.txt

@@ -30,6 +30,7 @@ compair_core/server/deps.py
 compair_core/server/settings.py
 compair_core/server/local_model/__init__.py
 compair_core/server/local_model/app.py
+compair_core/server/local_model/ocr.py
 compair_core/server/providers/__init__.py
 compair_core/server/providers/console_mailer.py
 compair_core/server/providers/contracts.py

{compair_core-0.3.15 → compair_core-0.4.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "compair-core"
-version = "0.3.15"
+version = "0.4.1"
 description = "Open-source foundation of the Compair collaboration platform."
 readme = "README.md"
 license = { text = "MIT" }

compair_core-0.3.15/compair_core/compair/feedback.py

@@ -1,204 +0,0 @@
-from __future__ import annotations
-
-import os
-from typing import Any, Iterable, List
-
-import requests
-
-from .logger import log_event
-from .models import Document, User
-
-try:
-    import openai  # type: ignore
-except ImportError:  # pragma: no cover - optional dependency
-    openai = None  # type: ignore
-
-try:
-    from compair_cloud.feedback import Reviewer as CloudReviewer  # type: ignore
-    from compair_cloud.feedback import get_feedback as cloud_get_feedback  # type: ignore
-except (ImportError, ModuleNotFoundError):
-    CloudReviewer = None  # type: ignore
-    cloud_get_feedback = None  # type: ignore
-
-
-class Reviewer:
-    """Edition-aware wrapper that selects a feedback provider based on configuration."""
-
-    def __init__(self) -> None:
-        self.edition = os.getenv("COMPAIR_EDITION", "core").lower()
-        self.provider = os.getenv("COMPAIR_GENERATION_PROVIDER", "local").lower()
-        self.length_map = {
-            "Brief": "1–2 short sentences",
-            "Detailed": "A couple short paragraphs",
-            "Verbose": "As thorough as reasonably possible without repeating information",
-        }
-
-        self._cloud_impl = None
-        self._openai_client = None
-        self.openai_model = os.getenv("COMPAIR_OPENAI_MODEL", "gpt-4o-mini")
-
-        if self.edition == "cloud" and CloudReviewer is not None:
-            self._cloud_impl = CloudReviewer()
-            self.provider = "cloud"
-        else:
-            if self.provider == "openai":
-                api_key = os.getenv("COMPAIR_OPENAI_API_KEY")
-                if api_key and openai is not None:
-                    # Support both legacy (ChatCompletion) and new SDKs
-                    if hasattr(openai, "api_key"):
-                        openai.api_key = api_key  # type: ignore[assignment]
-                    if hasattr(openai, "OpenAI"):
-                        try:  # pragma: no cover - optional runtime dependency
-                            self._openai_client = openai.OpenAI(api_key=api_key)  # type: ignore[attr-defined]
-                        except Exception:  # pragma: no cover - if instantiation fails
-                            self._openai_client = None
-                if self._openai_client is None and not hasattr(openai, "ChatCompletion"):
-                    log_event("openai_feedback_unavailable", reason="openai_library_missing")
-                    self.provider = "fallback"
-            if self.provider == "local":
-                self.model = os.getenv("COMPAIR_LOCAL_GENERATION_MODEL", "local-feedback")
-                base_url = os.getenv("COMPAIR_LOCAL_MODEL_URL", "http://local-model:9000")
-                route = os.getenv("COMPAIR_LOCAL_GENERATION_ROUTE", "/generate")
-                self.endpoint = f"{base_url.rstrip('/')}{route}"
-            else:
-                self.model = "external"
-                self.endpoint = None
-
-    @property
-    def is_cloud(self) -> bool:
-        return self._cloud_impl is not None
-
-
-def _reference_snippets(references: Iterable[Any], limit: int = 3) -> List[str]:
-    snippets: List[str] = []
-    for ref in references:
-        snippet = getattr(ref, "content", "") or ""
-        snippet = snippet.replace("\n", " ").strip()
-        if snippet:
-            snippets.append(snippet[:200])
-        if len(snippets) == limit:
-            break
-    return snippets
-
-
-def _fallback_feedback(text: str, references: list[Any]) -> str:
-    snippets = _reference_snippets(references)
-    if not snippets:
-        return "NONE"
-    joined = "; ".join(snippets)
-    return f"Consider aligning with these reference passages: {joined}"
-
-
-def _openai_feedback(
-    reviewer: Reviewer,
-    doc: Document,
-    text: str,
-    references: list[Any],
-    user: User,
-) -> str | None:
-    if openai is None:
-        return None
-    instruction = reviewer.length_map.get(user.preferred_feedback_length, "1–2 short sentences")
-    ref_text = "\n\n".join(_reference_snippets(references, limit=3))
-    messages = [
-        {
-            "role": "system",
-            "content": (
-                "You are Compair, an assistant that delivers concise, actionable feedback on a user's document. "
-                "Focus on clarity, cohesion, and usefulness."
-            ),
-        },
-        {
-            "role": "user",
-            "content": (
-                f"Document:\n{text}\n\nHelpful reference excerpts:\n{ref_text or 'None provided'}\n\n"
-                f"Respond with {instruction} that highlights the most valuable revision to make next."
-            ),
-        },
-    ]
-
-    try:
-        if reviewer._openai_client is not None and hasattr(reviewer._openai_client, "responses"):
-            response = reviewer._openai_client.responses.create(  # type: ignore[union-attr]
-                model=reviewer.openai_model,
-                input=messages,
-                max_output_tokens=256,
-            )
-            content = getattr(response, "output_text", None)
-            if not content and hasattr(response, "outputs"):
-                # Legacy compatibility: join content parts
-                parts = []
-                for item in getattr(response, "outputs", []):
-                    parts.extend(getattr(item, "content", []))
-                content = " ".join(getattr(part, "text", "") for part in parts)
-        elif hasattr(openai, "ChatCompletion"):
-            chat_response = openai.ChatCompletion.create(  # type: ignore[attr-defined]
-                model=reviewer.openai_model,
-                messages=messages,
-                temperature=0.3,
-                max_tokens=256,
-            )
-            content = (
-                chat_response["choices"][0]["message"]["content"].strip()  # type: ignore[index, assignment]
-            )
-        else:
-            content = None
-    except Exception as exc:  # pragma: no cover - network/API failure
-        log_event("openai_feedback_failed", error=str(exc))
-        content = None
-    if content:
-        content = content.strip()
-        if content:
-            return content
-    return None
-
-
-def _local_feedback(
-    reviewer: Reviewer,
-    text: str,
-    references: list[Any],
-    user: User,
-) -> str | None:
-    payload = {
-        "document": text,
-        "references": [getattr(ref, "content", "") for ref in references],
-        "length_instruction": reviewer.length_map.get(
-            user.preferred_feedback_length,
-            "1–2 short sentences",
-        ),
-    }
-
-    try:
-        response = requests.post(reviewer.endpoint, json=payload, timeout=30)
-        response.raise_for_status()
-        data = response.json()
-        feedback = data.get("feedback") or data.get("text")
-        if feedback:
-            return str(feedback).strip()
-    except Exception as exc:  # pragma: no cover - network failures stay graceful
-        log_event("local_feedback_failed", error=str(exc))
-
-    return None
-
-
-def get_feedback(
-    reviewer: Reviewer,
-    doc: Document,
-    text: str,
-    references: list[Any],
-    user: User,
-) -> str:
-    if reviewer.is_cloud and cloud_get_feedback is not None:
-        return cloud_get_feedback(reviewer._cloud_impl, doc, text, references, user)  # type: ignore[arg-type]
-
-    if reviewer.provider == "openai":
-        feedback = _openai_feedback(reviewer, doc, text, references, user)
-        if feedback:
-            return feedback
-
-    if reviewer.provider == "local" and getattr(reviewer, "endpoint", None):
-        feedback = _local_feedback(reviewer, text, references, user)
-        if feedback:
-            return feedback
-
-    return _fallback_feedback(text, references)