GallowGlass AI, focused on process automation, data governance, and tools that respect emerging AI norms, today announces Boswell 1.0, an open-source application for English-literature analysis: explore a curated corpus, retrieve passages with concordance and semantic search, curate an Evidence tray, and synthesize answers grounded in the excerpts you choose — with local LLMs and vector search under your control.
Built for students, teachers, and close readers
Boswell supports an evidence-first workflow: pick a literary period (with optional soft scope across neighboring eras), run concordance and semantic search over indexed works, add strong lines to the Evidence tray, then use Evidence chat and Synthesize to draft analysis tied to real quotations — not free-form hallucination. A Knowledge Base path ingests your documents when object storage is configured; Graph Lab embeds Memgraph Lab to visualize period–work relationships.
Example works in Boswell
The repository includes a bundled public-domain sample for demos and CI: Shakespeare's Sonnet 18 (docs/gutenberg/). Running init_data.py seeds periods and indexes that text into Memgraph and Qdrant alongside the starter corpus.
Boswell's corpus manifest (Project Gutenberg IDs) names representative works across English literary history:
| Period | Works |
|---|---|
| Old English | Beowulf (Hall tr.) |
| Early Modern | Shakespeare's Hamlet; Marlowe's The Tragical History of Doctor Faustus |
| Sacred / Liturgical | King James Bible; Book of Common Prayer (Scottish Episcopal edition) |
| Late Modern & Romantic | Coleridge's The Rime of the Ancient Mariner; Wordsworth & Coleridge's Lyrical Ballads (1798); Byron's Childe Harold's Pilgrimage; Boswell's The Life of Johnson |
| Victorian & Fin de Siècle | Tennyson's Idylls of the King; Dickens's A Tale of Two Cities, Oliver Twist; Conan Doyle's The Adventures of Sherlock Holmes, The Sign of the Four |
Key capabilities
- Corpus & period scope — Browse works and periods; header controls scope soft neighbors for borderline texts.
- Concordance & keywords — Keyword-in-context and frequency-style exploration within scope.
- Semantic search — Chunk embeddings in Qdrant, scoped via Memgraph (with SQL fallback) to eligible works.
- Evidence tray — Persistent working set of excerpts with stable locators for synthesis and citation.
- Evidence chat & Synthesize — Chat and drafting with tray-grounded context; verify claims against allowed excerpts.
- Knowledge Base — Uploads and ingest pipeline (Docling-supported formats where enabled); derived Markdown for previews.
- Graph Lab — Same-origin embed of Memgraph Lab for period–work exploration.
- Home & status — General chat on Home; app status surfaces
/healthand/health/readyfor dependencies.
Local-first analysis
Boswell is designed to run on your stack: Ollama on the host for embeddings and chat models, Postgres for works and metadata, Qdrant for vectors, Memgraph for relationship scoping. Prompts and corpus data stay in your environment when you deploy that way — suitable for coursework, labs, and air-gapped setups.
Stack
- Backend: Python 3.11+, FastAPI
- Data: PostgreSQL, Memgraph (Bolt / Cypher), Qdrant
- AI: Ollama on the host (
host.docker.internalfrom containers) - Frontend: React, Vite, TypeScript, Tailwind CSS, Shadcn-style UI
- Deployment: Docker Compose
Get started
Repository: github.com/byronichero/boswell
git clone https://github.com/byronichero/boswell.git
cd boswell
cp .env.example .env
docker compose up --buildInstall Ollama models (nomic-embed-text, llama3.2, or your choice), then run scripts/init_data.py in the backend container to seed periods, sample Gutenberg text, Memgraph, and Qdrant. See the repository README and docs/ guides (user, admin, developer).
Boswell takes its name from Samuel Johnson's faithful chronicler — your literary companion for inspectable, evidence-grounded reading.
Contact: ppk@gallowglass.ai