Impact Vision

Open-source AI-powered impact measurement and SDG alignment agent for VC and impact investment funds.

Built on OpenHarness, Impact Vision ships a conversational AI agent, a CLI, a REST API, an MCP server, a Streamlit dashboard, and a single-file Web Console — all backed by the same engine with deep expertise in GIIN's IRIS+ framework, UN SDGs, the 5 Dimensions of Impact, and 10+ ESG / regulatory frameworks (ISSB, ESRS, SFDR, TCFD, SASB, GRI, PCAF, SBTi, EU Taxonomy, TNFD, CDP).

Release history lives in CHANGELOG.md. Strategy and engineering plans live in docs/.

Screenshots

Try it yourself: See a sample HTML report generated for a pig farm in Malaysia.

What is Impact Investing?

Impact investing means investing with the intention to generate positive, measurable social and environmental impact alongside a financial return. Unlike traditional investing (financial return only) or philanthropy (social good only), impact investing seeks both.

Key concepts Impact Vision helps with:

Concept	What it means
IRIS+	The "GAAP for impact" -- ~787 standardized metrics for measuring social/environmental outcomes (maintained by GIIN)
SDGs	17 UN Sustainable Development Goals (e.g., No Poverty, Clean Energy, Climate Action) with 169 targets
5 Dimensions	The standard framework for assessing impact quality: What outcome? Who benefits? How much? Would it happen anyway? What could go wrong?
Impact DD	Due diligence focused on whether an investment will actually generate the claimed impact
ESG	Environmental, Social, Governance -- risk management frameworks (SASB, GRI, TCFD, SFDR, EDCI, UNPRI, ISSB, ESRS)
NESTA Evidence	5 levels rating how strong the evidence is: Level 1 (narrative) to Level 5 (rigorous RCT)

Core Use Case

Upload a pitch deck or investment memo and Impact Vision will:

1. Extract and classify impact claims (outcome / output / activity / intent / risk)
2. Map claims to relevant IRIS+ metrics from the 787-metric catalog
3. Detect SDG goal/target alignment from the content
4. Run an impact DD checklist (122 questions across 34 categories from GIIN, PCV, Seraf, IMP, AFME + sector-specific for 15 sectors)
5. Assess evidence strength using NESTA Standards of Evidence (levels 1-5)
6. Auto-extract a Company model for immediate use with downstream assessment tools
7. Compare against sector benchmarks from GIIN survey data
8. Suggest the most important follow-up questions for the investment team
9. Generate reports in HTML (with Plotly charts), XLSX, CSV, JSON, or text

Using Impact Vision in your fund workflow? See the Fund Manager Quick Reference for a Python-first 60-second SDK walkthrough (deal scoring, IC memo, DD questionnaire, portfolio roll-up, LP calendar).

Quick Start (from scratch)

1. Prerequisites

You need Python 3.11+ and Git. Check if they're installed:

python --version    # should show 3.11 or higher
git --version       # any recent version

If not installed: Python | Git

2. Clone and install

git clone https://github.com/joejoe168168/impact-vision.git
cd impact-vision

Create a virtual environment (recommended):

python -m venv .venv

# Activate it:
# Windows PowerShell:  .venv\Scripts\Activate.ps1
# Windows CMD:         .venv\Scripts\activate.bat
# Mac/Linux:           source .venv/bin/activate

Install the package:

pip install -e ".[dev]"

After this, you'll have two commands available: impact-vision and iv (shorthand).

'impact-vision' is not recognized? Run the auto-fix script:

# Windows PowerShell
powershell -ExecutionPolicy Bypass -File scripts\add-to-path.ps1

# Windows CMD
scripts\add-to-path.bat

# Mac/Linux
bash scripts/add-to-path.sh

Important: After running the script, you must close and reopen your terminal (CMD/PowerShell/Terminal) for the PATH change to take effect. Then try impact-vision --help.

Alternative: Use python -m openharness instead (works without PATH changes):

python -m openharness --help
python -m openharness catalog stats
python -m openharness dd list

3. Set up an LLM provider

The AI agent needs a language model. We recommend OpenRouter for beginners -- it's free to start and gives you access to many models.

Option A: OpenRouter (recommended -- free models available)

Step 1. Create an OpenRouter account:

1. Go to openrouter.ai and click Sign Up (Google or GitHub login works)
2. Go to openrouter.ai/keys and click Create Key
3. Copy the key (starts with sk-or-...)

Step 2. Run the interactive setup wizard:

impact-vision setup

The wizard will guide you through these prompts:

? Choose a provider workflow:
  Anthropic-Compatible API  Claude / Kimi / GLM / MiniMax
> OpenAI-Compatible API  OpenAI / OpenRouter          <-- select this
  ...

? Choose an OpenAI-compatible provider:
  OpenAI official
> OpenRouter                                          <-- select this

? Base URL: https://openrouter.ai/api/v1              <-- press Enter (default)
? Default model: openai/gpt-oss-120b:free             <-- type a model name

? Enter API key for OpenRouter: sk-or-your-key-here    <-- paste your key

Done! You can now start the agent (Step 4).

Free models (verified May 2026): Browse the live list at openrouter.ai/models?q=free. Impact analysis needs tool calling, so stick with these:

Model	Context	Why we recommend it
openai/gpt-oss-120b:free	131K	OpenAI open-weight MoE, native tool use, strong reasoning
nvidia/nemotron-3-super-120b-a12b:free	1M	1M-token window for long decks, strong multi-step reasoning
z-ai/glm-4.5-air:free	131K	Hybrid-thinking MoE, cheap + fast, tool-calling
google/gemma-4-31b-it:free	262K	Native function calling, multimodal, Apache 2.0

OpenRouter rotates free endpoints monthly — if a model 404s, re-run impact-vision setup and pick another from the live list.

Option B: Anthropic (Claude Sonnet) -- best quality for impact analysis

1. Go to console.anthropic.com, create an account
2. Go to API Keys and create a key
3. Run the wizard and choose Anthropic-Compatible API > Claude official:

impact-vision setup

1. Paste your API key when prompted (Claude Sonnet is the default model)

Option C: OpenAI (GPT-5)

1. Go to platform.openai.com/api-keys, create a key
2. Run the wizard and choose OpenAI-Compatible API > OpenAI official:

impact-vision setup

1. Paste your API key when prompted

Option D: Local Ollama (free, runs on your machine, no internet needed)

1. Install Ollama from ollama.com
2. Pull a model: ollama pull llama3.2
3. Run setup:

impact-vision ollama-setup --model llama3.2

No API key needed -- everything runs locally on your GPU/CPU.

Provider	Best for	Cost
OpenRouter	Trying multiple models, free tier	Free + pay-per-use
Anthropic (Claude Sonnet)	Best impact analysis quality	Pay-per-use
OpenAI (GPT-5)	General purpose	Pay-per-use
Ollama (local)	Privacy, offline use	Free (your hardware)

Model quality note: Impact analysis requires models that support tool calling (function calling). Free models vary in quality -- some may not follow the impact expert persona or use the analysis tools correctly. For best results:

• Recommended paid: Claude Sonnet or GPT-5 for production DD work
• Recommended free: openai/gpt-oss-120b:free, nvidia/nemotron-3-super-120b-a12b:free, z-ai/glm-4.5-air:free
• May struggle: Very small models (<9B) or models without tool-calling support

4. Start the AI agent

impact-vision

Try asking:

• "Analyze this pitch deck" (provide a path to a PDF)
• "What SDGs does a solar energy company align with?"
• "Run a 5-dimension assessment for a fintech serving 50,000 clients"

5. Try the CLI tools (no API key needed)

These commands work without any LLM setup:

# See all available commands
impact-vision --help

# Browse IRIS+ metrics (all 787 metrics included out of the box)
impact-vision catalog stats
impact-vision catalog search "climate"
impact-vision catalog search "gender"

# List ESG/sustainability frameworks
impact-vision framework list

# Quick multi-framework scan of a company description
impact-vision framework scan "Solar energy company providing clean power to 50,000 rural households"

# Cross-reference a metric across all frameworks
impact-vision framework xref OI4112

# Browse the Due Diligence checklist (122 questions across 34 categories)
impact-vision dd list
impact-vision dd categories

# Analyze text against the DD checklist
impact-vision dd analyze "We serve 45,000 clients across 3 countries. Our NPS score is 72."

6. Update the IRIS+ Catalog (optional)

All 787 IRIS+ 5.3c metrics are already bundled and work out of the box. If GIIN releases a newer version of the catalog:

1. Download the new Excel file from GIIN IRIS+ (free registration)
2. Place it in data/raw/
3. Run:

impact-vision catalog load    # Parse Excel into JSON cache
impact-vision catalog stats   # Verify metric count

7. Launch the dashboard (optional)

streamlit run src/openharness/dashboard/app.py

Opens a web dashboard at http://localhost:8501 with 5 tabs: Assessment, IRIS+ Catalog, DD Checklist, Framework Scan, and Portfolio.

Quick reference

Command	What it does
iv --help	Show all commands
iv catalog stats	Show catalog statistics
iv catalog search "query"	Search IRIS+ metrics
iv framework list	List all supported ESG / regulatory frameworks
iv framework scan "text"	Quick multi-framework scan
iv framework xref OI4112	Cross-reference a metric
iv dd list	Show all 122 DD questions
iv dd categories	List DD categories
iv dd analyze "text"	Check text against DD checklist
iv ollama-setup	Configure local LLM
iv serve-mcp	Start MCP server for AI agents (39 impact tools + 5 resources)
iv serve-web	Start Web Console + REST API at http://127.0.0.1:8787
iv	Start interactive AI agent

(iv is a shorthand for impact-vision)

Usage

Impact Vision ships 39 impact agent tools covering the full investment lifecycle — pre-screen, due diligence, IC memo, portfolio monitoring, LP reporting, assurance, and post-exit review. Below are common prompts you can paste into the interactive agent (or the Web Console at http://127.0.0.1:8787). The agent will pick the right tools for you.

Analyzing a Pitch Deck

The primary workflow: upload a PDF and let the agent analyze it.

> Analyze this pitch deck for impact: /path/to/pitch_deck.pdf

The agent will use the pitch_deck_analyze tool which:

• Extracts text from all PDF pages
• Identifies impact claims and maps them to IRIS+ metrics and SDGs
• Runs the full DD checklist against the document
• Presents addressed questions vs. gaps
• Suggests follow-up questions to ask the investment team

Follow-up prompts that chain the v3/v4 tools on the same deal:

> Run a 5-dimension assessment for BrightPath Finance with the claims you just extracted
> Draft a Theory of Change for BrightPath and link it to IRIS+ metrics
> Score greenwashing risk per-claim and show which claims need verification
> Start an engagement workspace for BrightPath at the "scoping" stage
> Attach a data pack: these metrics came from the audited FY24 statements
> Build a completeness scorecard for the data pack and produce coaching cards
> Run the AI extraction review queue over the claims you flagged as low-confidence

Browsing the IRIS+ Catalog

> Search for IRIS+ metrics related to financial inclusion
> Show me metrics mapped to SDG 7 (Clean Energy)
> What metrics are tagged with the "What" dimension?
> Get details for metric OI1479

Running a DD Checklist

> Show me the full impact DD checklist
> Which DD questions are about risk assessment?
> Analyze this document against the DD checklist: /path/to/memo.pdf

SDG Alignment Scoring

> Map BrightPath Finance to SDG goals. They report metrics PI4060, OI8869, OI6213...
> Which SDGs does this company align with based on their financial inclusion work?

5-Dimension Assessment

> Score this company on the 5 Dimensions of Impact
> What are the gaps in the "Contribution" dimension?

Cross-Reference Lookup

> Look up cross-references for IRIS+ metric OI4112
> Search cross-references for "gender" across all frameworks
> What's the GRI equivalent of SFDR PAI indicator #1?

Generating Reports

> Generate an HTML impact report for BrightPath Finance (includes Plotly charts)
> Export the assessment as XLSX for our LP report
> Generate a report with sector benchmark comparison

HTML reports include an Interactive Score Improvement section -- check boxes for practices your organization follows (e.g., "We track beneficiaries", "We have a Theory of Change") and watch scores update in real-time with a before/after radar chart.

Improving Scores Through Q&A

Ask the agent to help you improve your impact scores interactively:

> Help me improve my impact scores for this pig farm
> Ask me questions to strengthen the assessment

The agent will:

1. Identify your weakest scoring dimensions
2. Ask targeted questions (e.g., "How many direct beneficiaries?", "Do you track emissions?")
3. Map your answers to IRIS+ metrics
4. Re-run the assessment and show exactly how your answers improved each dimension

ESG Framework Assessment

> Scan this company against all ESG frameworks
> What are the material SASB topics for a fintech company?
> Assess TCFD alignment for our climate disclosure
> Check SFDR PAI compliance for the portfolio
> Show the EDCI 2026 private-markets KPI fields and which we're reporting
> Run a UNPRI self-assessment for our fund
> Assess ISSB IFRS S1 general disclosure readiness
> Check ISSB IFRS S2 climate disclosure for our carbon data
> Run EU CSRD/ESRS double materiality assessment
> Classify our fund under SFDR Article 6/8/9

Theory of Change Assessment

> Assess our fund's Theory of Change against RS Group's Blended Value principles
> Check our ToC against the GIIN IRIS+ checklist
> Help me develop a Theory of Change for our microfinance investment
> List all RS Group principles and GIIN ToC steps

Greenwashing & Compliance Checks

> Run a greenwashing check on this pitch deck
> Assess EU Green Claims Directive compliance
> Check UK FCA Anti-Greenwashing Rule alignment
> Compute the Green Authenticity Index and Cheap Talk Index for this report

Impact Verification & Product Passport

> Check our readiness for IFC OPIM verification
> Import Digital Product Passport data from this JSON file
> Map DPP categories to IRIS+ and ESRS metrics

LP DDQ Export

> Generate an ILPA DDQ response for BrightPath Finance
> Create a GIIN/IRIS+ impact report template
> Export EDCI annual survey as XLSX: output_path="edci_survey.xlsx"

Portfolio Batch Analysis

> Analyze this portfolio CSV file: examples/sample_portfolio.csv
> Generate aggregated SDG coverage for the portfolio
> Run portfolio roll-up with fund-level 5D scores
> Generate an LP report for our fund
> Show impact attribution by sector and geography

Pipeline Management

> Add EcoFinance to the pipeline at screening stage
> Transition EcoFinance to DD in progress with rationale "Strong SDG alignment"
> Show the pipeline dashboard
> List all companies at IC review stage

Continuous Monitoring

> Set quarterly monitoring for EcoFinance
> Record metric PI4060 = 15000 for EcoFinance
> Check alerts for our portfolio
> Run a full re-assessment for EcoFinance

Guided Assessment

> Start a screening assessment for BrightPath Finance
> What's the next step in the assessment?
> Submit company description data for the current step

Stakeholder Voice & Beneficiary Feedback

> Build a Lean Data survey template for smallholder farmers in Kenya
> Register GDPR-compliant consent records for 50 beneficiaries
> Score feedback quality and link responses to our outcome claims

Climate Accounting (Scope 1/2/3 + PCAF)

> Calculate a Scope 1/2 GHG inventory from this fuel + electricity data
> Apply emission factor catalog v2 with uncertainty bands
> Run PCAF financed-emissions attribution for the loan book
> Check SBTi 1.5 °C alignment

AI Extraction Review & Evidence Governance

> Show the AI extraction review queue for BrightPath
> Auto-approve claims above 0.85 confidence that cite audited sources
> Flag any claim without a source URL for human review
> Open a verification workspace for the assurer and share only approved evidence

LP Narrative & Q&A (grounded in verified data)

> Generate an LP quarterly narrative for the Inclusive Finance fund
> Answer LP question "what % of beneficiaries are women" using only approved data
> Export the Q&A transcript with citations for the LPAC meeting

Portfolio Natural-Language Query

> What was the average CO2e intensity across the climate portfolio last year?
> Top 5 companies by beneficiary reach, verified data only
> Compare SDG 5 coverage between Fund I and Fund II

Exit Impact Assessment (OPIM Principle 8)

> Score exit-impact durability for Solar Co with acquirer profile "strategic utility"
> List unmitigated risks and build a 12-month exit impact plan

Consultant Engagements (v4 workspace)

The v4 engagement suite is a single tool (engagement_suite) that covers scoping, data rooms, ToC/KPI design, reporting studios, training, public website output, and the three-pillar assurance bundle.

> Create an engagement workspace for "Acme Solar Advisory Q2" at scoping stage
> Build a proposal with a fixed-fee rate card and send it for client signature
> Design a Theory of Change + KPI tree and validate it against IRIS+
> Build a client data room, score completeness, and issue coaching cards
> Move the draft report from "in_review" to "published" and log the state change
> Issue a readiness badge once training modules + diagnostic are complete
> Sign the final assurance bundle (evidence graph + audit trail + workspace) with HMAC

Single-Prompt Mode

For CI/CD or scripting:

impact-vision -p "Search IRIS+ catalog for climate-related metrics"

CLI Reference

impact-vision (or the iv shorthand) exposes seven top-level subcommand groups plus three service commands. Run any command with --help for full flags.

# Interactive agent
impact-vision                              # Start interactive agent session
impact-vision -p "your prompt"             # Single prompt, then exit
impact-vision --model opus                 # Use a specific model

# Provider / auth setup
impact-vision setup                        # Interactive provider wizard (OpenRouter/Claude/OpenAI/Ollama)
impact-vision ollama-setup --model llama3.2
impact-vision provider list | use | add | edit | remove
impact-vision auth   login | status | logout | switch | copilot-login | codex-login | claude-login

# IRIS+ catalog
impact-vision catalog load [EXCEL_PATH] [--force]
impact-vision catalog stats
impact-vision catalog search "climate"

# ESG / sustainability frameworks
impact-vision framework list
impact-vision framework scan "company description"
impact-vision framework xref OI4112

# Due-diligence checklist (122 questions / 34 categories)
impact-vision dd list [--category "What (Outcomes)"]
impact-vision dd categories
impact-vision dd analyze "text or /path/to/doc.txt"

# Service surfaces
impact-vision serve-mcp                                  # MCP server (stdio)
impact-vision serve-mcp --transport sse --port 8765      # MCP over SSE
impact-vision serve-web                                  # Web Console + REST API (http://127.0.0.1:8787)

# Developer utilities
impact-vision mcp      list | add | remove               # Manage MCP server configs
impact-vision plugin   list | install | remove           # Manage entry-point plug-ins
impact-vision cron     list | add | remove | run         # Cron scheduler for background jobs

Architecture

impact-vision/
├── src/openharness/
│   ├── impact/                        # Impact measurement engine
│   │   │
│   │   │   # --- Core engine ---
│   │   ├── models.py                  # Pydantic: Metric, Company, Assessment, SDG, ImpactClaim
│   │   ├── catalog.py                 # IRIS+ 5.3c Excel ETL (263-column parser)
│   │   ├── database.py                # In-memory MetricStore (search/filter/stats)
│   │   ├── sdg_taxonomy.py            # 17 SDG Goals + 169 Targets reference
│   │   ├── five_dimensions.py         # What/Who/HowMuch/Contribution/Risk scoring
│   │   ├── sdg_mapper.py              # Per-goal SDG alignment scorer (0-100)
│   │   ├── gap_analysis.py            # Core Metric Set coverage analysis
│   │   ├── dd_checklist.py            # DD question engine + NESTA evidence scoring
│   │   ├── benchmarks.py              # Sector benchmarks for 18 sectors
│   │   ├── greenwashing.py            # Greenwashing detection (standard + Green Claims + FCA + NLP)
│   │   ├── risk_opportunity.py        # Risk/opportunity (likelihood × severity)
│   │   ├── storage.py                 # SQLite persistence for assessments
│   │   │
│   │   │   # --- Fund workflow (v0.8+) ---
│   │   ├── fund_thesis.py             # Fund impact thesis, IC gate, adverse thresholds
│   │   ├── ic_memo.py                 # IC memo rendering (MD/HTML/DOCX/PPTX)
│   │   ├── deal_gate.py               # Deal scorecard (pass/warn/fail gate)
│   │   ├── portfolio_rollup.py        # Capital-weighted portfolio roll-up
│   │   ├── lp_calendar.py             # 12-month LP reporting calendar
│   │   ├── tenancy.py                 # Multi-tenant + RBAC
│   │   ├── plugins.py                 # Entry-point plug-in discovery
│   │   ├── signed_feed.py             # Hash-chained LP report feed (HMAC)
│   │   ├── lp_portal.py               # ILPA-compatible LP portal
│   │   ├── marketplace.py             # Thesis marketplace (publish/subscribe)
│   │   │
│   │   │   # --- Scientific rigor + primary data (v0.14.0) ---
│   │   ├── extractors/                # Pluggable claim extractors (regex/LLM)
│   │   ├── toc_graph.py               # Theory-of-Change graph + Mermaid renderer
│   │   ├── counterfactual.py          # GIIN COMPASS additionality templates
│   │   ├── bayes.py · meta_analysis.py · spillover.py · sroi.py · causal.py
│   │   ├── geospatial.py · surveys.py · worker_voice.py · ecosystem_services.py
│   │   ├── registries.py              # Verra/Gold Standard/Puro/BioCredits
│   │   ├── returns.py                 # MOI + impact-adjusted IRR
│   │   ├── external_benchmarks.py     # GIIN Compass peer quartiles
│   │   ├── blended_finance.py         # IL-Loans, SOC/DIB, impact carry
│   │   ├── assurance.py · csrd_wizard.py · issb_reporting.py · soc2_checklist.py
│   │   ├── audit_trail.py             # Hash-chained lifecycle events
│   │   ├── i18n.py · fx.py · regulatory_packs.py · branding.py
│   │   │
│   │   │   # --- v2 institutional backbone (v0.13+) ---
│   │   ├── metric_records.py          # Canonical MetricRecord contract
│   │   ├── investee_collection.py     # Questionnaire schema + submission lifecycle
│   │   ├── climate_accounting.py      # Scope 1/2 GHG inventory
│   │   ├── evidence_graph.py          # Claim↔metric↔target↔evidence lineage
│   │   ├── standards_registry.py      # Versioned standards metadata
│   │   ├── roadmap_v2.py              # Collection / disclosure / assurance helpers
│   │   │
│   │   │   # --- v3 Trust Infrastructure (v0.15.0) ---
│   │   ├── emission_factors.py        # Versioned factors + sensitivity bands
│   │   ├── stakeholder_voice.py       # Lean Data + GDPR/PDPA consent
│   │   ├── evidence_workflow.py       # AI extraction review queue
│   │   ├── verification_workspace.py  # Assurer workspace + findings
│   │   ├── lp_narrative.py            # LP narrative + Q&A (approved-data only)
│   │   ├── greenwashing_reviewer.py   # Per-claim explainable review
│   │   ├── portfolio_nlq.py           # NL portfolio queries + ApprovedDataPolicy
│   │   ├── exit_impact.py             # OPIM P7 exit-impact scoring + P8 learning context
│   │   │
│   │   │   # --- v4 Engagement Suite (latest) ---
│   │   ├── engagements/
│   │   │   ├── workspace.py           # EngagementWorkspace + artifact audit hook
│   │   │   ├── proposal.py            # Proposal builder + e-signature
│   │   │   ├── toc_builder.py         # Wraps toc_graph + metric_recommender
│   │   │   ├── data_room.py           # Completeness scorecard + coaching cards
│   │   │   ├── value_creation.py      # Scenario + business case + risk scoring
│   │   │   ├── reporting_studio.py    # Draft → review → published state machine
│   │   │   ├── training.py            # Modules + diagnostic + readiness badge
│   │   │   ├── website.py             # Public diagnostic + lead capture
│   │   │   ├── copilot.py             # Governed AI: review queue + safe answer
│   │   │   ├── regulatory.py          # Jurisdiction profiles + SFDR classification
│   │   │   └── verification_bundle.py # 3-pillar signed assurance bundle (HMAC)
│   │   │
│   │   ├── report_templates/          # Jinja2-based HTML report templates
│   │   ├── frameworks/                # 10 ESG/sustainability frameworks
│   │   │   ├── sasb.py · gri.py · tcfd.py · sfdr_pai.py · edci.py
│   │   │   ├── unpri.py · theory_of_change.py · issb_ifrs_s1.py · issb_ifrs_s2.py
│   │   │   ├── esrs.py · ifc_opim.py · pcaf.py · sbti.py · eu_taxonomy.py
│   │   │   ├── tnfd.py · cdp.py
│   │   │   └── cross_reference.py     # 59 cross-framework metric mappings
│   │   ├── mcp_server.py              # MCP server (FastMCP)
│   │   └── sdk.py                     # High-level ImpactVision SDK facade
│   │
│   ├── tools/impact/                  # 39 LLM-callable impact agent tools (see "Tools" below)
│   ├── api_gateway/router.py          # FastAPI REST API
│   ├── web/                           # Single-file Web Console + SSE streaming
│   ├── dashboard/app.py               # Streamlit 5-tab dashboard
│   ├── skills/bundled/content/        # Agent knowledge (markdown)
│   ├── prompts/system_prompt.py       # Impact Vision persona + instructions
│   └── cli.py                         # CLI (7 subcommand groups + serve-mcp / serve-web)
├── data/
│   ├── raw/                           # IRIS+ Excel file (not committed)
│   ├── processed/                     # JSON catalog cache (auto-generated)
│   ├── dd_checklist.yaml              # 122 DD questions / 34 categories
│   ├── scoring_config.yaml            # Sector baselines + keyword boosts
│   ├── sdg_keywords.yaml              # SDG keyword mappings for 20+ sectors
│   ├── core_metric_set_per_sdg.yaml   # Curated SDG core metric set
│   ├── fund_thesis.*.yaml             # Default + 4 regional thesis packs
│   └── i18n/                          # 6 languages (en/es/fr/pt/zh/ar)
├── docs/
│   ├── fund-manager-guide.md          # Python SDK walkthrough for funds
│   ├── roadmap-v3.md / -v3-implementation.md
│   ├── roadmap-v4.md                  # Consultant-led engagement suite
│   └── cursor-integration.md          # Cursor/VS Code MCP setup
├── examples/                          # Sample company, portfolio, MCP configs
├── tests/                             # Test suite (impact + v2 + v3 + v4)
└── .github/workflows/ci.yml           # Import smoke + tests + ruff

DD Checklist

The built-in due diligence checklist includes 122 questions across 34 categories, sourced from:

• GIIN Impact Toolkit - The Impact Due Diligence Guide
• Pacific Community Ventures - Impact DD Emerging Best Practices
• Seraf Toolbox - Impact Investing Due Diligence Checklist
• Impact Management Project (IMP) - Five Dimensions of Impact
• AFME / Neotas / OECD - ESG Due Diligence frameworks
• Sector-specific: 15 sectors including fintech, healthcare, agriculture, energy, education, manufacturing, transport, construction, tourism, retail, mining, media, professional services, waste management, and ICT

Each addressed question is assessed using NESTA Standards of Evidence (levels 1-5):

Level	Description
1	Narrative / anecdotal only (self-reported, no data)
2	Output data (quantified activities, e.g. # served)
3	Outcome data measured (pre/post, surveys, tracked KPIs)
4	Controlled comparison (quasi-experimental, benchmarks)
5	Rigorous evaluation (RCT, independent audit, causal attribution)

Questions are organized into 34 categories (18 core + 15 sector-specific + 1 SDG):

Category	Qs	Covers
Impact Thesis & Theory of Change	4	Mission, theory of change, business model alignment
What (Outcomes)	4	Specific outcomes, positive/negative impacts, outcome data
Who (Stakeholders)	5	Beneficiaries, underserved status, baseline, feedback
How Much (Scale)	5	Reach, depth, duration, growth, quantitative metrics
Contribution	4	Additionality, counterfactual, evidence, investor contribution
Risk	5	Evidence risk, execution risk, external risk, mitigation, impact washing
Measurement Systems	6	IMM systems, IRIS+ alignment, data frequency, third-party audit
Governance/ESG	5	Board oversight, environmental/labor/ethics policies, incentives
SDG Alignment	1	Specific goals and targets
Negative Impact	3	Do-no-harm assessment, grievance mechanisms
Exit Sustainability	3	Impact continuity, mission lock, acquirer risk
Financial Sustainability	5	Revenue model, impact-return tension, grants dependency, pricing
Team & Capability	4	Founder experience, community ties, key person risk
Market & Context	4	Market size, regulation, systemic barriers, competition
Product/Service Design	4	User-centered design, safety, privacy, affordability
Supply Chain	3	ESG practices, forced/child labor risk, environmental footprint
Stakeholder Voice	3	Feedback mechanisms, co-design, transparency
Investor Alignment	3	Impact covenants, value-add beyond capital, portfolio fit

Sector	Qs	Covers
Fintech	5	Over-indebtedness, client protection, effective interest rate, digital literacy, responsible AI
Healthcare	5	Health regulations, patient safety, clinical efficacy, affordability, data privacy
Agriculture	5	Farmer income, sustainable farming, climate resilience, food safety, land tenure
Energy	5	CO2e avoided, energy access, e-waste, affordability, grid reliability
Education	5	Learning outcomes, underserved learners, pedagogy, digital safety, employment
Manufacturing	2	Circular economy, pollution prevention, worker safety
Transport & Logistics	2	Emissions reduction, last-mile accessibility
Construction	3	Green building standards, affordable housing, waste diversion
Tourism	2	Cultural preservation, community benefit-sharing
Retail	3	Ethical sourcing, plastic waste reduction, fair labor
Mining & Extractives	3	Tailings management, community consent (FPIC), rehabilitation
Media	2	Misinformation safeguards, digital inclusion
Professional Services	2	Pro-bono access, diversity metrics
Waste Management	3	Recycling rates, informal worker integration, hazardous waste
ICT	3	E-waste, data sovereignty, digital divide

Frameworks & Standards

All frameworks below are exposed via the framework_assess tool, the MCP server, the REST API, and the Python SDK. Every framework ships with cross-references to IRIS+ metric IDs via the shared cross_reference module (59 concept mappings).

Category	Framework	Coverage
Core taxonomy	GIIN IRIS+ 5.3c	~787 metrics, SDG mappings, 5-Dimension tags
	UN SDGs	17 Goals, 169 Targets
	Impact DD Checklist	122 questions / 34 categories (GIIN, PCV, Seraf, IMP, AFME + 15 sectors) with NESTA evidence (1-5)
	Sector Benchmarks	18 sectors (GIIN survey data) with aggregated 5D scores and coverage
	Cross-Reference Mapping	59 concepts mapped across IRIS+/GRI/EDCI/SFDR PAI/SASB/TCFD/ESRS/ISSB/PCAF/SBTi/TNFD/CDP/EU Taxonomy
ESG disclosure	SASB	17 industries, 77+ material topics
	GRI	34 standards (Universal + Topic), 120+ disclosures
	TCFD / IFRS S2	4 pillars, 11 disclosures, scenario analysis, Scope 1/2/3
	EDCI	2026 PE/VC KPI fields, including non-core cybersecurity testing
	UNPRI	6 Principles, 27 actions
	Theory of Change	RS Group 8 Blended Value Principles + GIIN 8-step ToC Checklist
	ISSB IFRS S1	General sustainability disclosure (4 pillars)
	ISSB IFRS S2	Climate-related disclosures
	EU CSRD / ESRS	11 standards, double-materiality assessment
Regulatory	SFDR	14 mandatory + 9 optional PAI indicators, Article 6/8/9 classification, deadline scheduler
	EU Taxonomy	6 environmental objectives, DNSH + Minimum Safeguards
	UK FCA Anti-Greenwashing Rule	Fair/clear/not-misleading assessment
	EU Green Claims Directive	Evidence, comparability, third-party verification
	EU Digital Product Passport (ESPR)	Import + map to IRIS+/ESRS/SDG
	Per-jurisdiction packs	EU-SFDR, EU-CSRD, UK-FCA-SDR, US-SEC-ESG, HK-HKEX-ESG, AU-AASB-S2, ISSB-global
Climate & nature	PCAF	Financed-emissions attribution, sector defaults, weighted data quality
	SBTi (Net-Zero Standard v1.2)	1.5 °C pathway, Scope-3 materiality, 2050 cap
	TNFD v1	14 LEAP / pillar disclosures
	CDP	Climate / water / forests questionnaire intake
	GHG Protocol	Scope 1/2 inventory (Scope 3 via PCAF) with versioned factor catalog
Impact management	IFC OPIM	9-principle verification readiness + Principle 8 exit-impact
	SROI	Deadweight / attribution / displacement / drop-off adjustments
	MOI + Impact-adjusted IRR	Newton-Raphson, optional shadow price
Greenwashing & NLP	Standard greenwashing scoring	Vague-language + quantitative-evidence checks
	Green Authenticity Index (GAI)	Ratio of substantive to vague claims
	Cheap Talk Index (CTI)	Forward-looking vs. evidenced statements
	Per-claim explainable reviewer	concrete / mixed / vague / buzzword_only classification + severity
Assurance	ISAE 3000 / AA1000	Management-assertion + subject-matter + evidence register
	SOC 2 Type II / ISO 27001	Starter control set with readiness report
	Verification workspace	Finding lifecycle + threaded comments (v0.15.0)
	3-pillar assurance bundle	HMAC-signed evidence graph + audit trail + workspace (v4)

Agent Tools (39)

All tools below are exposed through the default OpenHarness tool registry and openharness.tools.impact, so the interactive agent, Web Console, REST API, and MCP server see the same surface.

Pre-screen & core assessment (7)

Tool	Description
pitch_deck_analyze	PDF/TXT/MD intake with impact-claim extraction + Company model
iris_catalog	IRIS+ catalog search, browse, filter by SDG/theme
sdg_mapper	SDG alignment scoring with theme inference and evidence chains
five_dimension_assess	5-Dimension assessment with additionality & counterfactual prompts
gap_analysis	Metric gap analysis vs Core Metric Set
impact_metric_recommender	Recommend IRIS+ metrics by theme, SDG, and sector
impact_data_quality	Quality score for reported metrics (placeholders, unknown IDs)

Due diligence & evidence (5)

Tool	Description
dd_checklist	122-question DD checklist, document analysis, and targeted suggestions
document_analysis	Multi-document comparison, change detection, claim verification
guided_assessment	Step-by-step workflow with deal-stage templates
verification_prep	IFC OPIM 9-principle readiness assessment
product_passport	EU Digital Product Passport import and IRIS+/ESRS mapping

Risk & credibility (4)

Tool	Description
greenwashing_detect	Composite screen (5 sub-scores) + Green Claims / FCA / GAI / CTI
greenwashing_reviewer	Per-claim explainable review with severity + governance metadata
impact_risk_opportunity	14 risk categories on a likelihood × severity matrix
exclusion_screening	UNGC, weapons, fossil-fuel exclusion lists

Frameworks & reporting (4)

Tool	Description
framework_assess	Multi-framework ESG assessment (all frameworks in the table above)
cross_reference	Cross-framework metric lookup (59 mappings)
impact_report	Interactive HTML reports + XLSX/CSV/JSON/text/PDF
lp_ddq_export	LP DDQ responses in ILPA, GIIN, EDCI, SFDR formats

Decision workflow — v5 (2)

Tool	Description
decision_workflow	Quick screen, IC memo proof bundle, deal comparison, and LP readiness
regulatory_calendar	Jurisdiction-specific reporting deadlines for fund and engagement planning

Portfolio workflow (5)

Tool	Description
portfolio_analyze	Batch analysis, capital-weighted roll-ups, benchmarking
portfolio_query	Natural-language portfolio queries (ApprovedDataPolicy-gated)
pipeline	8-stage investment pipeline with transition tracking
monitoring	Continuous monitoring, metric updates, alerts, re-assessment
trend_analysis	Time-series metric trend analysis with trajectory projection

Stakeholder voice & narrative (4)

Tool	Description
beneficiary_feedback	Import and analyze beneficiary feedback data
stakeholder_voice	Lean Data templates + GDPR/PDPA consent + feedback↔claim links
improvement_advisor	LLM-guided improvement recs, peer insights, SDG opportunities
narrative	Impact narrative drafting (exec summary, key findings, case studies)

Trust infrastructure — v3 (4)

Tool	Description
emission_factors	Versioned factor catalog, sensitivity bands, inventory repricing
evidence_review	AI extraction review queue with policy-driven auto-approval
verification_workspace	Assurer workspace with finding lifecycle and threaded comments
lp_narrative	LP narrative + Q&A constrained to verified data with citations

Exit & assurance (1)

Tool	Description
exit_impact	OPIM Principle 8 scoring + exit plan

Consultant engagement suite — v4 (3)

Tool	Description
engagement_workspace	Engagement lifecycle (scoping → delivery → closeout) + artifact audit
toc_builder	ToC canvas + logic-chain validator + multi-framework KPI generator (wraps v3 toc_graph + metric_recommender)
engagement_suite	Umbrella tool for Tracks 3-10: proposal, data room, value-creation, reporting studio, training/readiness, public website, governed AI copilot, regulatory deadlines, 3-pillar assurance bundle

Streamlit Dashboard

For a visual alternative to the CLI agent:

streamlit run src/openharness/dashboard/app.py

The dashboard has 5 tabs:

1. Company Assessment: Input company data, see 5-Dimension radar chart, SDG bar chart, and gap analysis
2. IRIS+ Catalog: Browse, search, and filter the 787-metric catalog
3. DD Checklist: Browse questions, paste text to check coverage
4. Framework Scan: Run TCFD, SFDR PAI, EDCI, and SASB assessments
5. Portfolio: Upload CSV for batch analysis with aggregated charts

Web Console (power-user UI)

For a browser-native surface to every tool — useful when you want the full 39-tool set at your fingertips rather than Streamlit's 5 curated tabs — run the web console:

# Start the console + REST API (defaults to http://127.0.0.1:8787)
impact-vision serve-web

# Or directly via uvicorn
uvicorn openharness.web.app:app --host 127.0.0.1 --port 8787

The console is a single self-contained HTML file (no build step, no JS framework) that sits on top of the existing FastAPI gateway:

• Lists every impact tool in a searchable sidebar (Ctrl/⌘+K to focus).
• Derives typed forms from /openapi.json and POSTs to /api/v1/*.
• Shows JSON results in a syntax-highlighted pane, with copy / save buttons.
• Persists every run to localStorage so you can re-hydrate old invocations.
• Optional bearer-token box for IMPACT_VISION_API_KEY-protected deployments.
• Links to the live /docs (Swagger) and the GitHub repo.

Development

# Install with dev dependencies
pip install -e ".[dev]"

# Run all tests
python -m pytest tests/ -v

# Focused subsets
python -m pytest tests/test_impact.py -v                    # engine + frameworks
python -m pytest tests/test_v4_tracks_3_to_10.py -v         # v4 engagement suite

# Import smoke checks (verifies all package exports work)
python scripts/check_imports.py --all

# Lint
ruff check src/

GitHub Actions runs import smoke, full tests, and ruff on every push/PR.

MCP Server (Use with Claude, Cursor, VS Code)

Impact Vision can run as an MCP server, exposing the full impact tool surface and 5 read-only resources to any MCP-compatible AI client.

impact-vision serve-mcp                              # stdio (desktop clients)
impact-vision serve-mcp --transport sse --port 8765  # SSE (remote clients)

Use stdio for local desktop clients such as Claude Desktop, Cursor, and VS Code. Use SSE when the MCP server is started separately and clients connect over HTTP.

Cursor / VS Code Setup

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "impact-vision": {
      "command": "impact-vision",
      "args": ["serve-mcp"]
    }
  }
}

Claude Desktop Setup

Copy examples/claude_desktop_config.json to your Claude Desktop config directory.

MCP Resources

Resource URI	Purpose
impact://catalog/stats	IRIS+ catalog counts, categories, and themes
impact://dd-checklist/categories	DD checklist categories and question counts
impact://frameworks/list	Supported ESG / impact frameworks
impact://cross-reference/{metric_id}	Cross-framework mapping for one metric
impact://sdg/goals	UN SDG goal reference data

See docs/cursor-integration.md for the full setup guide and client-specific notes.

REST API

FastAPI REST gateway backing the Web Console, MCP server, and third-party integrations:

# Start the API server
uvicorn openharness.api_gateway.router:app --reload

# Authenticated (set env var for production)
IMPACT_VISION_API_KEY=your-secret-key uvicorn openharness.api_gateway.router:app

Key endpoints: /api/v1/score, /api/v1/sdg-map, /api/v1/greenwashing, /api/v1/report, /api/v1/pipeline, /api/v1/batch, and more. See the auto-generated OpenAPI docs at /docs.

Roadmap

Strategy and engineering plans live in docs/:

• docs/roadmap-v2.md — April 2026 institutional-readiness plan (data contracts, investee collection, climate accounting, LP reporting, assurance controls, causal impact, governed AI).
• docs/roadmap-v3.md / -v3-implementation.md — Trust infrastructure (evidence review, verification workspace, LP narrative, portfolio NLQ, exit impact). Shipped.
• docs/roadmap-v4.md — Consultant-led engagement suite (Tracks 3-10). Backend shipped; frontend + paid-data wiring deferred to Wave 5.
• ROADMAP.md — historical engineering record.
• CHANGELOG.md — release notes.

Contributing

Have ideas? Open an issue or submit a PR!

License

MIT License. See LICENSE for details.

Acknowledgments

• AvantFaire Investment Management -- the first impact investment company in Hong Kong that nurtured the creator's passion for impact measurement
• GIIN for IRIS+ and the Impact Due Diligence Guide
• Pacific Community Ventures for DD emerging best practices
• Seraf for the impact investing DD checklist
• Impact Management Project for the 5 Dimensions of Impact
• OpenHarness for the agent infrastructure