Seedy-Research

Seedy Research

Thai-published research, connected to the world—and to the next study.

Seedy helps researchers find Thai evidence, read lawful full papers, compare it with global research, and turn reviewed findings into a reusable research plan. Universities and research institutes are the primary audience. Civil engineering is the first proof vertical, not the product boundary.

Try Seedy · Evaluator guide · Connect an agent · Run the checks

Public guest view captured September 4, 2026 (Asia/Bangkok); no mocked responses.

Research Preview. Implemented capabilities, dated verification, and expansion plans are distinguished below. Research output is not professional advice or proof of scientific novelty.

One research case, from question to next study

1. Discover: start with a question and search research published in Thailand. Inspect a dated visibility receipt: exact global identity, metadata gaps, candidate match, no exact match in that audit, not audited, or unavailable.
2. Read: open the cited page or rights-cleared native full paper. Keep the original source, version, license, and page locator beside the finding.
3. Review in Workspace: screen selected papers, compare methods and findings in a literature-review matrix, and record PRISMA-guided decisions.
4. Synthesize in Notebook: send reviewed Workspace findings into the same Research Case. Use Sources–Chat–Studio for source-grounded conversation, persistent notes, and versioned research artifacts.
5. Connect and plan: inspect exact-DOI OpenAlex relationships, build a Research Path, and frame a candidate gap and falsifiable Next-Study Protocol.
6. Review and export: reopen the exact pages and accept or reject claims in a Research Passport. Export preserves evidence, metadata, and inference as separate layers.

Workspace and Notebook are distinct, connected features. Notebook Light Mode uses bounded retrieval and model APIs over existing source records; the heavier OpenRAG adapter is optional and inactive by default. This is not a claim that the OpenRAG server is deployed.

Why Seedy, alongside OpenAlex?

OpenAlex provides the global scholarly graph. Seedy works on the local evidence and research workflow around it: Thai-provider discovery, dated visibility comparisons, lawful page inspection, human review, and reusable research artifacts. It does not submit or repair records in OpenAlex.

A failed lookup is not proof that a paper is globally invisible. Global relationships remain metadata-only leads until the underlying evidence is independently available and reviewed. A generated gap is a candidate to test, not proven novelty.

WebMCP: the agent and researcher share the page

Twelve browser-native tools reuse the application's APIs and update the same visible Case, evidence drawer, connection map, Passport, Research Path, Workspace, and Notebook.

Tool	Purpose
start_research_case	Start or resume a question and bounded Thai discovery
discover_research	Find Thai-local sources and optional global comparison leads
audit_global_visibility	Inspect a dated comparison receipt
inspect_paper_evidence	Open bounded exact-page evidence and lawful reader state
trace_research_connections	Trace an exact-DOI identity into global metadata
draft_research_passport	Draft from inspected anchors, preserving the review gate
build_research_path	Create or adapt an evidence-grounded study path
inspect_learning_progress	Inspect checkpoints without exposing private answers
open_research_notebook	Open the visible Case Notebook and list public Case Sources
send_reviewed_to_notebook	Continue human-reviewed Workspace evidence as a versioned pack
ask_research_notebook	Create a cited answer from explicit public Case Sources
draft_notebook_artifact	Draft a cited, versioned Studio artifact for human review

Tools validate inputs and honor cancellation. Source text is untrusted input. Browser agents cannot turn metadata into evidence, read private Notebook history, verify Workspace cells for a person, or bypass Passport review. The separate remote MCP service supports clients without an open browser page.

Data and verification

September 2, 2026 snapshot:

Cohort	Recorded scope
ThaiJO discovery catalog, including native papers	2,681 records
Thai-published native full papers	103 papers
Thai-affiliated global comparison full papers	897 papers
Legacy civil page-citable corpus	1,297 papers

The native reader total is 1,000 papers / 14,485 pages; it is not 1,000 Thai-local papers. TNRR, TCI, TDC, conferences, and institutional repositories have separate promotion and partnership gates. National completeness is not claimed. Dated corpus details and integrity limits · Source rights · Expansion plan

The repository includes three rights-reviewed reader papers (68 pages) and synthetic test fixtures. The production corpus is not distributed in Git. Fixture tests verify application contracts, not live retrieval quality, authenticated database isolation, host compatibility, or traffic capacity. See verification scopes and commands and rollout requirements.

Reproduce without production credentials

Requires Python 3.10, Node.js 20, and npm. From a fresh clone:

git clone https://github.com/LeChiffreVol2/Seedy_Research.git
cd Seedy_Research
python3.10 -m venv .venv310
. .venv310/bin/activate
python -m pip install 'beautifulsoup4>=4.12,<5' 'requests>=2.32,<3'
npm ci --prefix web
make fixture-check
cd web && npx playwright install chromium && cd ..
make fixture-browser

No Supabase project or paid API key is needed for this fixture gate. It runs real application code with explicitly mocked external boundaries; it is not a standalone offline research service. For live local services, database setup, and provider keys, follow Operations.

Repository map

• web/ — Next.js interface, server routes, and browser-native SeedyMCP tools.
• mcp-server/ — FastAPI retrieval and remote MCP contracts.
• pipeline/ — bounded harvesting, rights-aware ingestion, and visibility audits.
• supabase/ — schema and additive migration history.
• harness/, eval/ — contract, retrieval, security, and release checks.

Architecture · Product thesis · Compatibility and archived work

License and responsible use

The source code is MIT licensed. That license does not grant rights to papers, extracted text, or third-party datasets. Private uploads remain owner-scoped and are not added to the public corpus. Privacy · Report a source or support issue