OpenGP is a self-hosted, multi-tenant intelligence platform for private-equity firms, fund managers, and family offices — an open-source alternative to Rogo, Hebbia, and BlueFlame. Every AI answer is grounded in your documents with clickable citations, and everything runs in your own AWS account: your data never leaves your cloud.
- Cited AI chat (RAG) — ask questions across your data room and portfolio; every figure links to the exact passage it came from. Click a citation, see the source page.
- Data room — fund/transaction document management with server-side parsing (OCR + tables), chunking, embeddings, and hybrid retrieval (pgvector + full-text + rerank).
- Portfolio analytics — deal explorer, MOIC/IRR/DPI KPIs, sector and geography breakdowns, and an AI chart builder over your live portfolio data.
- Model Builder — describe an analysis ("model portfolio MOIC if sector weights shift") and it writes and runs real code in an isolated sandbox, returning editable Excel models.
- Transaction reports — cited deal summaries generated from the data room, with custom report types.
- Meeting intelligence — a meeting bot joins your calls, transcribes, detects questions live, and answers them grounded in your fund data.
- LP inbox — email a question to your workspace's inbox address and get a cited draft answer; LP emails get suggested replies (the AI never emails LPs directly).
- News & outreach — tracked-entity news with LLM relevance classification, plus LinkedIn outreach orchestration.
- Multi-tenant & white-label — workspaces with per-workspace branding, roles (admin / analyst / viewer), and Postgres row-level security separating every tenant.
| Cited answers | Meeting live-assist |
|---|---|
 |
 |
| Model Builder | Data room |
|---|---|
 |
 |
This is not a demo app — it's the production architecture we ran commercially, open-sourced:
- Aurora PostgreSQL (Serverless v2) with pgvector + full-text search, ~96 row-level-security policies enforcing tenant isolation at the database layer
- Amazon Bedrock for document parsing (Bedrock Data Automation), embeddings (Cohere Embed v4), and reranking (Cohere Rerank 3.5) — plus bring-your-own model keys per workspace, routed per feature (chat, reports, transcription, news) via a KMS-encrypted key vault
- Cognito authentication with TOTP MFA, KMS encryption, CloudTrail audit trail, private S3 buckets with server-side key reconstruction (no presigned-key confusion)
- The entire stack is Terraform (
infra/terraform/) — ~77 resources, deployable into your own account with four variables - CI on every PR: build + tests (2,200+), dependency audit, secret scan, SAST, and a gate that bans direct DB access from the browser
QUICKSTART.md walks zero-to-running in ~half a day — including the first-workspace bootstrap script. The short version:
git clone https://github.com/vivan1211/opengp.git
cd opengp && npm install && npm test # everything runs locally except AWS services
cd infra/terraform
cp backend.hcl.example backend.hcl # point at your state bucket
terraform init -backend-config=backend.hcl
terraform plan # review!
terraform applyThen follow docs/DEPLOYMENT.md to build the Docker images (CodeBuild →
ECR) and roll them onto the app server, and docs/aws-migration/AWS-SETUP.md
for the full service-by-service reference. Database schema lives in
supabase/migrations/ (plain SQL, applied with psql; the directory name
is historical).
| Doc | What's in it |
|---|---|
| QUICKSTART.md | Zero to running in half a day, in order |
| DEVELOPMENT.md | Monorepo layout, build/test commands, the gotchas |
| docs/DEPLOYMENT.md | Deploy runbook + pre-deploy security checklist |
| docs/aws-migration/AWS-SETUP.md | Every AWS service, end to end |
| infra/terraform/README.md | The IaC: what's codified, what's hand-operated, how to apply |
| docs/PAUSE-AND-RESUME.md | Take the stack offline / bring it back (save money while idle) |
| docs/aws-migration/CLOUDFLARE-EDGE.md | Optional Cloudflare edge (built, documented, currently off) |
In-app, the Product info pages (/changelog) document the architecture and security model
interactively — the architecture diagram above is a screenshot of it.
Tools like Rogo, Hebbia, and BlueFlame AI prove the demand for AI-native workflows in private markets — but they're closed SaaS: your fund data lives on their infrastructure, priced per seat. OpenGP is the same category of product with the opposite deployment model: AGPL-3.0 licensed, self-hosted in your own AWS account, with your own model keys. Fork it, brand it, extend it.
OpenGP was built as a commercial platform for a European private-equity workspace and open-sourced in August 2026 with all customer data, branding, and identifiers removed (the git history begins at the open-source release for that reason). It shipped to production; it is no longer actively developed by the original author. Issues and PRs are welcome — serious fork/maintainer interest doubly so.
AGPL-3.0. You can use, modify, and self-host OpenGP freely — including commercially. If you offer it to others as a service, the AGPL requires you to share your modifications.