Repository navigation
Releases
Releases · ernaneDomingues/noul-search-agent
Noul v2.2.1
What's Changed
- ci: bump the actions group across 1 directory with 5 updates by @dependabot[bot] in #56
- fix(search): bound search inputs; a repeated source runs once by @ernaneDomingues in #60
- fix(net): connect-time address checks, bounded decoding, download deadline, key scope by @ernaneDomingues in #61
- fix: treat third-party text as inert everywhere it leaves Noul by @ernaneDomingues in #62
- fix(export): third-party metadata cannot break out of exported files by @ernaneDomingues in #63
- fix(workspace): tighter local file handling for paths a model may be steered into by @ernaneDomingues in #64
- fix(server): fairer and bounded limits on the public connector by @ernaneDomingues in #65
- ci: least privilege and pinned inputs in every workflow and the image by @ernaneDomingues in #66
- chore(release): 2.2.1 by @ernaneDomingues in #69
New Contributors
- @dependabot[bot] made their first contribution in #56
Full Changelog: v2.2.0...v2.2.1
Noul v2.2.0
What's Changed
- chore: English-only repository, clean-up and Noul brand on the site by @ernaneDomingues in #47
- fix: revive a dead test, isolate three test modules and remove CLI/site drift by @ernaneDomingues in #48
- refactor(tests): share the test helpers and cover the citation graph tools by @ernaneDomingues in #49
- refactor: shared workspace, cache, endpoint and annotation helpers by @ernaneDomingues in #50
- refactor: one builder for the paper record; lookups tell missing from unavailable by @ernaneDomingues in #51
- refactor: split the tool modules into packages and long functions into named steps by @ernaneDomingues in #52
- refactor: shared MCP/REST dispatch, CLI command table, installer client table, generated manifests by @ernaneDomingues in #53
- chore: stricter gates (ruff RUF100/SIM/C90/BLE, mypy check_untyped_defs, narrowed excepts, coverage 88) by @ernaneDomingues in #54
- fix(server): add an open /health route; Cloud Run never forwards /healthz by @ernaneDomingues in #57
- chore(release): 2.2.0, first release on PyPI by @ernaneDomingues in #58
- docs(site): publish the public connector URL by @ernaneDomingues in #59
Full Changelog: v2.1.0...v2.2.0
Noul v2.1.0
What's Changed
- fix(release): publish the GitHub Release when signing jobs are skipped by @ernaneDomingues in #41
- refactor(v2): Phase 4a — retractions and preprint <-> published links by @ernaneDomingues in #42
- refactor(v2): Phase 4b — CSL-JSON, project exports and zotero_push by @ernaneDomingues in #43
- refactor(v2): Phase 4c — project archive/delete, logged searches and 2.1.0 by @ernaneDomingues in #44
- fix(tests): close sqlite connections so the store test passes on Windows by @ernaneDomingues in #45
- fix(skill): report instructions planted in untrusted text at the top of the answer by @ernaneDomingues in #46
Full Changelog: v2.0.0...v2.1.0
Noul v2.0.1
Full Changelog: v2.0.0...v2.0.1
v1.0.0 — Autonomous Scientific Research & Document Intelligence Agent
Noul Search Agent v1.0.0 — Autonomous Scientific Research & Document Intelligence Agent ("Co-Scientist")
First stable production release (v1.0.0) of Noul Search Agent, delivering all 6 Roadmap Phases (Issues #1 to #28) with 41 MCP tools, 29 CLI subcommands, zero mandatory external dependencies, and 155/155 automated unit tests passing in both Pydantic v2 and pure-stdlib fallback modes.
- Documentation & Interactive Portal: https://ernanedomingues.github.io/noul-search-agent/
- Repository: https://github.com/ernaneDomingues/noul-search-agent
Highlights by Capability
1. Multi-Disciplinary Discovery, Scope Clarification & Artifact Radar (Phase 1 — #1, #2, #3, #19)
- 6 Scientific Literature Sources:
OpenAlex,PubMed,Semantic Scholar,Crossref,arXiv, andEurope PMC+UnpaywallOpen Access resolution. - Deterministic Evidence Meter (Noul Meter): Raw and study-design-weighted synthesis (
meta-analysis->preclinical). - Scientific Domain Classifier (
detect_paper_domain/noul domain): Automatic classification acrosscomputer_science,biomedical,economics, andgeneral_stem. - Open Artifact Radar (
scan_paper_artifacts/noul artifacts): Detects and verifies live GitHub/GitLab code, HuggingFace/Zenodo weights and datasets, and flags evasive availability statements ("available upon reasonable request"). - Influential Citation Graph (
get_paper_references/noul references): Walks backward and forward citations ranked by Semantic ScholarinfluentialCitationCountwith OpenAlex fallback. - Interactive Scope Clarification (
clarify_search_scope/noul clarify): Disambiguates broad queries before searching.
2. Multi-Paper Synthesis, PKM & Query Expansion (Phase 2 — #4, #5, #6, #7)
- Comparative Matrix (
compare_papers/noul compare): Side-by-side Markdown comparison across methodology, sample/dataset, metrics, and limitations for 2–10 papers. - Thematic Insights (
get_thematic_insights/noul insights): Extracts cited consensuses, methodological tensions, and future work directions. - Obsidian PKM Export (
export_obsidian_notes/noul export --format obsidian): Generates YAML frontmatter,[[Wikilinks]], and BibTeX notes confined to the workspace. - Boolean Query Expansion (
suggest_boolean_queries/noul expand): Suggests 3 structuredAND/OR/NOTqueries (broad,balanced,strict).
3. Technical Deep-Dive & Claim Grounding (Phase 3 — #8, #9, #10, #11, #12, #13)
-
Benchmark Table Extractor (
extract_benchmark_tables/noul benchmarks): Normalizes Markdown, LaTeX, and plain-text tables and identifiessota_winner. -
Hardware & VRAM Estimator (
estimate_hardware_specs/noul hardware): ComputesFP16/BF16,INT8, andINT4VRAM + KV Cache for Dense and MoE models. -
Strict Claim Grounding (
verify_paper_claims/noul ground): Verifies quantitative/comparative claims with exact page/table/section attributions ([pág. X, Tabela Y: "..."]). -
Math & Algorithm Transpiler (
extract_paper_math/noul math): Extracts equations and pseudocode as clean LaTeX andast.parse-validated typed Python functions. -
Ablation Impact Analyzer (
analyze_ablation_tables/noul ablation): Ranks components by$|\Delta%|$ and flags marginal (< 1%) or dominant (>= 60%) components. -
Prompt & Decoding Extractor (
extract_prompts_and_decoding/noul prompts): Captures prompt templates with verbatim placeholders, decoding hyperparameters, and JSON schemas.
4. Failure Modes & Temporal Preprint Auditing (Phase 4 — #14, #15)
- Failure Modes & Assumptions Detector (
detect_failure_modes/noul failures): Classifies failure modes (long_context,noise_sensitivity,out_of_distribution,latency_or_compute,sample_or_population_bias), critical assumptions, and scenarios where baselines outperform the proposed method. - arXiv Version Diff Analyzer (
diff_arxiv_versions/noul diff): Comparesv1vsv2+snapshots to track metric deltas, claim attenuations/strengthenings, and newly added baselines.
5. Domain-Adaptive Multi-Persona Peer Review, Polyglot Scaffolding & Hypothesis Engine (Phase 5 — #16, #17, #18, #20, #21)
-
Multi-Persona Peer Review Committee (
review_paper/noul review): Specialized 3-reviewer + Meta-Reviewer committees tailored toCS/AI(NeurIPS/ICLR),Biomedical(CONSORT/STROBE/PRISMA), andEconomics(Causal Identification, IV/DiD/RDD & Robustness). -
Polyglot Code & Protocol Scaffolder (
scaffold_implementation/noul scaffold): Generatesast.parse-validatedPyTorch(nn.Module+ tensor shape unit tests),Statsmodels,R(fixest), orClinical Protocolscaffolds. -
Autonomous Research Gap & Hypothesis Engine (
generate_research_hypotheses/noul hypotheses): Cross-analyzes$\ge 2$ papers to formulate testable hypotheses with proposed methodology, datasets, and validation metrics.
6. Academic Authoring Copilot, BibTeX Sanitizer, Rebuttal & Pre-Submission Gatekeeper (Phase 6 — #22, #23, #24, #25, #26, #27, #28)
- Contextual Citation Infiller (
recommend_citations/noul cite): Detects[CITATION NEEDED],\cite{?}, and unbacked empirical claims, retrieving supporting literature with justifications and@articleBibTeX entries. - IMRaD Blueprint Generator & Draft Auditor (
generate_manuscript_blueprint,audit_manuscript_draft/noul draft structure|audit): Generates venue-specific IMRaD templates and audits.md/.texdrafts for weak claims, notation inconsistencies, and reproducibility omissions. - Related Work Synthesizer (
synthesize_related_work/noul related-work): Groups bibliography by approach families into comparative academic prose + gap positioning paragraph in Markdown or LaTeX. - BibTeX Sanitizer & Published Venue Resolver (
sanitize_bibtex/noul bibtex sanitize --update-preprints): Upgrades arXiv preprints to peer-reviewed conference/journal entries (@inproceedings/@article), deduplicates keys, and fixes LaTeX escaping. - Rebuttal Copilot (
parse_and_draft_rebuttal/noul rebuttal parse): Segments reviewer comments, flags new experiment requests, and drafts structured Markdown/LaTeX rebuttals. - Pre-Submission Desk-Reject Gatekeeper (
audit_submission_readiness/noul gatekeeper audit): Audits double-blind anonymity violations (non-anonymous GitHub URLs, author metadata, first-person self-citations), venue checklists, overclaiming, and mandatory sections, emitting a0–100Readiness Score.
Quick Start
# Run directly with uvx or npx
uvx noul-search-agent install --all
npx -y noul-search-agent install --all
# Or from source (Python >= 3.10, zero mandatory dependencies)
python3 install.py --all