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 Bedrockfor document parsing (Bedrock Data Automation), embeddings (Cohere Embed v4), and reranking (Cohere Rerank 3.5) — plus bring-your-own model keysper workspace, routed per feature (chat, reports, transcription, news) via a KMS-encrypted key vault Cognitoauthentication with TOTP MFA, KMSencryption, CloudTrailaudit trail, private S3buckets with server-side key reconstruction (no presigned-key confusion) The entire stack is Terraform**() — ~77 resources, deployable into your own account with four variablesinfra/terraform/
- 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 apply
Then follow ** docs/DEPLOYMENT.md** to build the Docker images (CodeBuild → ECR) and roll them onto the app server, and
for the full service-by-service reference. Database schema lives in
docs/aws-migration/AWS-SETUP.md(plain SQL, applied with psql; the directory name is historical).
supabase/migrations/
| Doc | What's in it |
|---|---|
DEVELOPMENT.mddocs/DEPLOYMENT.mddocs/aws-migration/AWS-SETUP.mdinfra/terraform/README.mddocs/-AND-RESUME.mddocs/aws-migration/CLOUDFLARE-EDGE.mdIn-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.