Vespera

Local-first AI deal analysis and due diligence.

Read a dataroom, cross-check its claims, score it against your investment thesis, and get an indicative valuation range — without your documents ever leaving your machine.

pip install vespera
vespera review ./dataroom --thesis my-thesis.md

Vespera reads the documents in a local folder — contracts, board minutes, financials, NDAs — and produces a structured deal report. All analysis runs on your machine via Ollama. Document contents never leave your network: no cloud upload, no third-party AI provider, no account.

What it does

Built for the first-pass review in M&A, VC, and PE deals:

• Due diligence findings — change-of-control clauses, termination rights, exclusivity, IP ownership, material liabilities, missing signatures — each with severity, a verbatim evidence excerpt, and the source file and page
• Mechanically verified citations — the source file and page on every finding are stamped by code, not by the model, and every evidence quote is checked in code against the source document; a quote that cannot be matched verbatim is labelled as inference, so you never have to take a citation on trust
• Run record — every report states the model, version, date, and verification counts behind it, and deal.json holds the full machine-readable analysis, so any figure can be explained months later
• Key metrics — revenue, ARR, growth, margins, retention, runway, extracted only where explicitly stated, every value cited to its source
• Contradiction detection — the same metric reported differently in two documents, conflicting claims across contracts and board minutes, referenced schedules that aren't in the dataroom
• AI adoption profile — is AI evidenced in the product, operations, and engineering, or only claimed? "AI-powered" marketing with a rule-based mechanism underneath is flagged as a red flag, and the profile feeds the valuation: AI-native economics, AI-augmented operations, and AI-adoption headroom carry different margin structures and multiples
• Deal readiness score — a reproducible severity-weighted score with a Strong / Balanced / Cautious reading
• Thesis fit — write your investment thesis once in Markdown; every deal is scored against it, with aligned points, conflicts, and unknowns
• Indicative valuation — a multiples-based screening range with every assumption listed (a range to interrogate, never an appraisal)

Vespera

Reviewing ./dataroom

Documents found: 9

Deal readiness: 44/100 — Balanced reading
AI posture: AI claimed, not evidenced
Indicative range: 38.4–76.8m GBP (screening only)
Thesis fit: 55/100

Findings:
- Inconsistencies between documents: 3
- Termination rights: 4
- Change-of-control clauses: 1
- Missing signatures: 1

Report: vespera-output/report.md
Evidence: vespera-output/findings.json · vespera-output/deal.json

All document analysis was performed locally.

Use it as a library too — the full analysis is one function returning one typed object:

from pathlib import Path
from vespera.deal import analyze_dataroom

analysis = analyze_dataroom(Path("./dataroom"), thesis_path=Path("my-thesis.md"))
print(analysis.score.score, analysis.score.label)

Privacy model

• No cloud calls for analysis. Inference runs on a local Ollama server (localhost by default).
• No telemetry, no accounts, no database. Vespera reads your documents and writes two output files. That's it.
• The only network activity you'll ever need is ollama pull to download a model once.

Quick start

1. Install Ollama — the free app that runs AI models on your own computer — and open it once.

2. Install Vespera (needs Python 3.12+):

   pip install vespera

3. Point it at a folder of documents (optionally with your thesis):

   vespera review ./dataroom --thesis my-thesis.md

That's it. On the first run Vespera automatically downloads its default local model (qwen3:4b, ~2.6 GB, one-time) and then starts the review. If anything is missing, Vespera tells you exactly what to do.

Prefer a more thorough (slower) review? Use --model qwen3:8b. See what's available with vespera models.

Try it on the included synthetic example:

git clone https://github.com/VesperaSystems/vespera
cd vespera
vespera review ./examples/sample-dataroom --thesis ./examples/thesis.md

Commands

vespera review PATH [--thesis thesis.md] [--model qwen3:8b] [--output vespera-output] [--host http://localhost:11434]
vespera models      # show recommended + locally installed Ollama models
vespera --version

The thesis file is plain Markdown — write your criteria however you normally would (see examples/thesis.md).

Supported document types

Format	Notes
PDF	Priority format; per-page source references
DOCX	Paragraphs and tables; no page numbers
TXT / MD	Plain text

Scanned image-only documents are not analysed in this version (no OCR).

Limitations

Vespera is automated document triage. It is not legal, financial, or investment advice, and it does not replace review by qualified professionals. The indicative valuation is a multiples-based screening range resting entirely on stated assumptions — it is not an appraisal and must not be relied on for any decision. Local language models can miss issues and misread context; findings must be verified against the source documents. Vespera is designed to tell a human professional where to look first — not to make decisions.

Roadmap

• IC memo generation from your firm's own template
• More valuation approaches (DCF, VC method) with the same assumptions-first framing
• OCR for scanned documents
• More document formats (XLSX, EML, PPTX)
• Additional local model providers (llama.cpp, MLX)
• Multi-language document support

Development

git clone https://github.com/VesperaSystems/vespera
cd vespera
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
pytest

The LLM is behind a tiny provider interface (vespera/llm/base.py), so tests inject a fake provider and never require Ollama.

License

Apache-2.0 — Copyright 2026 Daniel Molloy