Investors and financial analysts
Portfolio screening
Compare disclosure quality, transition plans, and source evidence across companies using your own sustainability criteria.
Open core AI toolkit
Sustainability report analysis you can verify
Move from one-off AI prompts to structured analysis with traceable source passages, configurable assessment frameworks, and repeatable quality checks.
Try OSA now!From general-purpose AI to a professional workflow
Already using ChatGPT or Claude for sustainability reports?
Tools like ChatGPT or Claude can answer questions from a report. Professional analysis also needs to find the relevant passages, show how each answer was produced, and give human analysts enough transparency to review and sign off the results. OSA adds that structure without replacing expert judgement.
Who it is for
Auditors, assurance teams, and consultants
Disclosure review
Review ESRS, CSRD, TCFD, and client-specific disclosures with traceable sources and a method that can be reused.
Sustainability, procurement, and research teams
Benchmarking and supplier risk
Benchmark reports, assess climate, nature and human-rights disclosures, or build transparent research datasets.
Move beyond individual prompts
Use your own questions, retrieve the relevant passages, review the evidence, and export results ready for analyst sign-off.
Features
From PDF upload to reproducible evaluation, OSA keeps the full analysis workflow under your control.
Upload one report or many
Analyze a single sustainability report or batch-process a collection of PDFs.
Use any question framework
Start with TCFD and ClimRetrieve question sets, or bring your own YAML framework.
Control every analysis
Choose the model, processing stage, chunk size, overlap, retrieval depth, and scoring.
Export and compare
Download answers and chunks as structured data, then compare reports and configurations.
See it in action
Inspect the workflow, not just the final answer
Configure the assessment, review its analysis and selected chunks, verify the highlighted source evidence, and calculate performance against a benchmark.
Core
Select framework, model, and parameters
Choose an assessment framework, model, and processing stage. Set chunk size, overlap, retrieval depth, and scoring before the run.
Core
See comprehensive analysis results
Review scored answers beside their source chunks. Evidence selections stay visible, and both the analysis and chunks can be exported.
Enterprise
Find and verify sources in the PDF
Move from an evidence chunk to the exact PDF page. The matching passage is outlined, making every finding easier to inspect and defend.
Core
Calculate assessment performance
Evaluate configurations against expert annotations with MAP, MRR, precision, recall, F1, and NDCG instead of relying on intuition.
Open core, built to inspect
The analysis engine is open source — you can audit every step, extend the pipeline, and run it on your own terms. Enterprise features like API access, managed search backends, and job processing are available separately. Developed with research partners to keep methods reproducible and aligned with current ESG standards.
Research collaboration with SRN
We work with the Sustainability Reporting Navigator — an academic platform by researchers at LMU Munich, University of Cologne, and others — to integrate expert-labeled sustainability data and benchmarks with OSA's open analysis pipelines.
Read the announcementStep-by-Step Deployment & Setup
Run sustainability report analysis with AI on Render or Streamlit — on your laptop, in the cloud, or fully managed. New here? Follow a step-by-step setup guide.
On-premise with Docker
Install locally with a single Docker command. Your data stays on your infrastructure.
One-click cloud
Deploy to your own cloud installation in one click. No credit card required.
Custom & supported
Need a tailored set-up, integrations, or ongoing support? We work with you to get it right.
Get step-by-step guidance for a managed cloud deployment.
Try OSA now!Research and funding partners
Part of the Climate+Tech Open Sustainability Analysis Framework. Developed and validated with leading universities, research labs, and funding organizations.
Claude integration
Using Claude for sustainability report analysis?
Open Sustainability Analyst can integrate with Claude and add what a chat workflow does not provide consistently: structured, explainable answers linked to report evidence.
A validation sprint compares your pipeline's answers and selected passages with expert-labelled examples, so you can measure quality before scaling it.
Book a validation sprintApplied work
Case studies
How teams use OSA to structure sustainability analysis, validate evidence, and improve specialised AI pipelines.
Case study · Actwyser
Making Claude quality measurable
“We were already using Claude for sustainability report analysis, but needed a more reliable way to validate results. With Climate+Tech, we turned our prompt-based workflow into a structured OSA prototype with clear question sets, source validation, and a ground-truth dataset. We can now see where AI helps, where it fails, and how to build a more robust analysis process.”
How we use OSA
As a platform to ensure that our sustainability report analysis meets our quality criteria.
What we solved
Structured workflow
Converted Claude answer policies into OSA question sets and a private prototype.
Ground-truth validation
Compared answer quality, source accuracy, and agreement with analyst judgement.
Source-based quality checks
Checked whether the model found the right passages and where stricter review was needed.
Case study · Score4More + University of Zurich
Retrieving useful evidence, not just relevant text
“We worked with Climate+Tech and University of Zurich researchers to improve how our AI pipeline selects evidence from sustainability reports. By combining real analyst questions, expert labels, and model fine-tuning, we could measure whether a passage was not only relevant, but useful for answering the question.”
What we achieved
Expert-labelled benchmark
Three professional analysts produced 1,061 gold annotations across 15 reports and 64 questions.
Pipeline evaluation
Compared retrieval, embedding, and LLM methods on relevance and decision-usefulness.
Fine-tuning findings
Improved usefulness classification while identifying a trade-off in model calibration.
Custom model development
Train a custom open-source LLM for your needs
Turn your expert criteria and labelled examples into a model for your domain. Climate+Tech helps define the task, prepare the training data, fine-tune the model, and evaluate it against a benchmark before deployment—building on the approach developed with Score4More and University of Zurich.
Discuss a custom modelGet started
Deploy Open Sustainability Analyst, connect with the community, or get step-by-step guidance for enterprise installation.
Contact
Questions about Open Sustainability Analyst, enterprise installation, or research collaboration? Get in touch.