Add arXiv Omi integration app

[wly12312]

Summary

• add a standalone no-auth arXiv integration app under plugins/omi-arxiv-app
• expose Omi chat tools for paper search, paper metadata lookup, and recent author papers
• include Railway/Heroku deployment files and local usage docs

Validation

• python -m py_compile plugins/omi-arxiv-app/main.py
• git diff --check
• FastAPI TestClient manifest checks for all 3 tools and type: object schemas
• helper checks for string/invalid limits, arXiv paper ID cleanup, category validation, and manifest schemas
• live arXiv smoke check for search_papers

Candidate integration app for #3120.

[greptile-apps]

Greptile Summary

This PR adds a self-contained arXiv integration plugin under plugins/omi-arxiv-app that exposes three Omi chat tools — paper search, paper detail lookup, and author search — backed by the public arXiv Atom API with no auth or environment variables required.

• main.py: FastAPI app with a shared httpx.AsyncClient managed by a lifespan context, input sanitization helpers (_safe_paper_id, _safe_category, _safe_limit), Atom feed XML parsing, and a /.well-known/omi-tools.json manifest endpoint.
• Deployment files (Procfile, railway.toml, runtime.txt): ready for Railway or Heroku, pinning Python 3.11.9 and starting uvicorn on $PORT.
• requirements.txt: fully pinned dependency set (fastapi, uvicorn, pydantic, httpx) with no conflicts.

Confidence Score: 4/5

The plugin is a new standalone directory with no changes to the main backend; the only risk is in the arXiv integration logic itself, which is well-isolated.

The code is clean and well-structured. The two issues found — double XML traversal in _format_entry and silent rejection of old-style versioned arXiv IDs like hep-th/9905001v1 in _safe_paper_id — are both minor and non-blocking for the common case. No auth, secrets, or database access is involved.

plugins/omi-arxiv-app/main.py — specifically the _safe_paper_id version-stripping logic and _format_entry's double call to _entry_authors.

Important Files Changed

Filename	Overview
plugins/omi-arxiv-app/main.py	Core FastAPI app with three Omi chat tools backed by the arXiv Atom API; minor issues with double XML traversal in _format_entry and old-style versioned IDs silently rejected by _safe_paper_id
plugins/omi-arxiv-app/requirements.txt	Pins fastapi, uvicorn, pydantic, and httpx to specific versions; compatible set with no obvious conflicts
plugins/omi-arxiv-app/Procfile	Standard Heroku/Railway Procfile; starts uvicorn on $PORT with a sensible default of 8080
plugins/omi-arxiv-app/railway.toml	Railway deployment config with NIXPACKS builder, /health check, and ON_FAILURE restart policy
plugins/omi-arxiv-app/runtime.txt	Pins Python 3.11.9 for Heroku/Railway runtime
plugins/omi-arxiv-app/README.md	Clear local dev and deployment docs with working curl examples for all three tools
plugins/omi-arxiv-app/.gitignore	Standard Python gitignore for .venv, pycache, and .pyc files

Sequence Diagram

sequenceDiagram
    participant Omi as Omi Client
    participant App as omi-arxiv-app (FastAPI)
    participant arXiv as arXiv Atom API

    Omi->>App: GET /.well-known/omi-tools.json
    App-->>Omi: tool manifest (search_papers, get_paper_details, search_author)

    Omi->>App: POST /tools/search_papers
    App->>App: _build_search_query() sanitize inputs
    App->>arXiv: "GET /api/query?search_query=...&max_results=N"
    arXiv-->>App: Atom XML feed
    App->>App: _parse_entries() + _format_entry()
    App-->>Omi: "ChatToolResponse {result}"

    Omi->>App: POST /tools/get_paper_details
    App->>App: "_safe_paper_id() validate & clean"
    App->>arXiv: "GET /api/query?id_list=XXXX&max_results=1"
    arXiv-->>App: Atom XML feed
    App-->>Omi: "ChatToolResponse {result}"

    Omi->>App: POST /tools/search_author
    App->>arXiv: "GET /api/query?search_query=au:Name&sortBy=submittedDate"
    arXiv-->>App: Atom XML feed
    App-->>Omi: "ChatToolResponse {result}"

Loading

_{Reviews (1): Last reviewed commit: "Add arXiv Omi integration app" | Re-trigger Greptile}

[greptile-apps]

_entry_authors is called twice on lines 158–160, traversing the XML tree twice. Storing the result in a variable avoids the redundant walk.

Suggested change

	authors = ", ".join(_entry_authors(entry)[:5]) or "unknown authors"
	if len(_entry_authors(entry)) > 5:
	authors += ", et al."
	all_authors = _entry_authors(entry)
	authors = ", ".join(all_authors[:5]) or "unknown authors"
	if len(all_authors) > 5:
	authors += ", et al."

[wly12312]

Addressed in 9f1991b: _entry_authors(entry) is now stored once in all_authors and reused for formatting.

[greptile-apps]

Old-style versioned arXiv IDs such as hep-th/9905001v1 or cs/9901001v2 pass through the version-strip branch unchanged (it only fires for new-format YYMM.NNNNvN IDs), so fullmatch on \d{7} fails and _safe_paper_id returns None. Users who paste a versioned legacy ID will get a confusing "invalid paper ID" error. Extend the split to cover old-format IDs as well.

Suggested change

	candidate = candidate.split("v", 1)[0] if re.match(r"^\d{4}\.\d{4,5}v\d+$", candidate) else candidate
	candidate = candidate.split("v", 1)[0] if re.match(r"^\d{4}\.\d{4,5}v\d+$", candidate) or re.match(r"^[a-z\-]+(\.[A-Z]{2})?/\d{7}v\d+$", candidate) else candidate

[wly12312]

Addressed in 9f1991b: _safe_paper_id now strips versions for both new-format and legacy arXiv IDs such as hep-th/9905001v1.

[greptile-apps]

No rate-limit protection for arXiv API calls

arXiv's usage policy asks automated clients to stay under 3 requests per second. With no rate-limiting or retry-with-backoff logic here, concurrent users could collectively trigger 429/503 responses that surface as opaque httpx.HTTPError messages. Consider adding a simple delay between requests or at minimum returning a user-friendly message when arXiv returns a 429.

[wly12312]

Addressed in 9f1991b: arXiv requests are serialized with a small delay to stay below 3 req/s, and 429/503 responses now return user-friendly messages.