1 Ethical AI sourcing
We build on open weights models (DeepSeek, GLM, Mistral) from labs without military, surveillance, or advertising contracts. No Big Tech dependencies, no opaque algorithms.
AI-assisted tools for nonprofits to enhance their advocacy, research, and communication.
Helping nonprofits find evidence, check every quotation against its source, and stress-test reports before publication. With ethical, open weights models instead of Big Tech AI.
4 tools for people who write evidence-based reports. Each covers one stage of a report, and they share one rule: a model may point at text, but only code extracts it.
Clear standards for responsible AI in nonprofit work.
We build on open weights models (DeepSeek, GLM, Mistral) from labs without military, surveillance, or advertising contracts. No Big Tech dependencies, no opaque algorithms.
Your data stays yours. We only work with providers that have zero-data-retention contracts and don't train on your content. The Enterprise plan enables self-hosted deployment, on your servers with your own model keys.
Our tools provide feedback and analysis to help you produce better, more rigorous work. We focus on augmenting nonprofit expertise, not replacing it: we don't facilitate the generation of low-effort content, grant applications, or full reports.
AI has limits: it can hallucinate, reflect biases, and miss context. Because of this, our tools have stringent, non-AI, deterministic controls. All output that is unverifiable is labelled, and we encourage human review and validation for all critical outputs.
We built AdvocacyLab because we needed better research, review, and analysis tools for our own advocacy work. Our evaluation framework and benchmark are how we ensure it delivers.
A structured methodology for assessing AI models across ethics, security, privacy, and fitness for nonprofit work. Every model and capability in AdvocacyLab is vetted through this framework.
See our AI evaluation training →The first AI benchmark designed for nonprofit advocacy, testing models' ideological positioning, bias detection, source and evidence assessment, and sensitive content handling. Results directly inform model selection and capability development.
What is live, and what we would build next. Features are rolled out based on nonprofit feedback and funding.
Says what the people a report criticises are likely to reply, and where it disagrees with itself.
Checks every quotation in a finished report against its source.
Verify claims in reports against trusted sources and knowledge bases, so evidence is accurate and well supported.
Turn survey results, programme metrics and other data into draft outlines, visualisations and key messages.
Turns a research question into passages cut word for word from their sources.
A team collects a report's quotations, each checked against its source as it is added.
Search databases of previous reports, your organisation's or others', for relevant information, precedents and evidence.
Ground the tools' answers in your organisation's past work and the sources it trusts.
Upload data files and get recommendations on statistical methods, key performance indicators and visualisation.
For data science teams: a review of your methodology, suggested improvements and help interpreting complex results.
Our approach versus generic AI tools
We've worked in social and humanitarian sectors for years. We understand nonprofit needs, constraints, workflows, and ethical considerations.
Built on open weights AI models and zero-data-retention providers, from labs without military, surveillance, or advertising contracts. We focus on strengthening human work, not generating low-effort content or automating core nonprofit functions.
One set of tools from research to publication: finding sources, collecting quotations while writing, and checking the finished report. One subscription instead of several.
Plans for individuals and teams, with an Enterprise plan priced by the size of your organisation.
Your data stays yours. We use zero-data-retention model providers, don't sell your data, and offer optional self-hosted deployment for full control.
AdvocacyLab aligns with funder priorities: it strengthens nonprofit capacity, improves advocacy quality, and ensures evidence-based approaches. We focus on review, feedback, and analysis. Not on generating content.
4 tools (CiteCheck, CiteCollect, ReportReview, and CiteFind) are live. We're seeking support to build the next tools on the roadmap.
Get in touchConnect with us to find out how we can help you use data and AI for greater impact.
The initial consultation is always free.
Drop us an email ✉️