Rereflect

---

What is Rereflect?

Rereflect ingests customer feedback from CSV, email, webhooks, Slack, Intercom and Zendesk, syncs with issue trackers like Linear, Jira and Asana, then uses NLP and (optionally) an LLM to classify sentiment, surface pain points and feature requests, flag urgent churn risks, and route everything through a team workflow — all behind a multi-tenant dashboard you host yourself.

• 🔓 100% open source (MIT). No "open core", no locked features.
• 🏠 Self-hosted — your data never leaves your box. Ships with Docker Compose.
• 🛰️ Zero telemetry. No usage analytics, no crash reporting, no phone-home. A default install makes no outbound calls of its own — only the LLM/integration endpoints you configure yourself. See Telemetry.
• 🔑 Bring your own LLM key. OpenAI, Anthropic or Google, encrypted at rest. There is no vendor key and nothing is proxied.
• 💸 Free by default. Runs end-to-end on a local VADER + keyword pipeline with no API key and zero cost. Add a key only when you want LLM-grade analysis.
• ✅ Everything unlocked. No tiers, seat caps, or feedback quotas — advanced churn, cohorts, analytics, integrations and the API are all included.

Highlights

🧠 AI feedback analysis	Sentiment, pain points, feature requests, urgency and topic clustering — local (VADER default) or LLM-powered (BYOK). Optional opt-in local transformer sentiment model (CPU, offline, per-org) with a built-in accuracy card that compares it against VADER on a labeled set — honest numbers, off by default. Per-org self-improving classifiers — separate sentiment, category and urgency heads train on your own corrections and auto-promote only when measurably better (≥ +0.02 macro-F1), fully reversible. The urgency head is add-only in auto (it can escalate to urgent, never silently clears the flag).
💬 AI Copilot (Cmd+K)	Ask questions in natural language, generate reports and draft responses — including recurring scheduled reports with optional email delivery (BYO-key Resend). When an answer returns a customer list, the copilot can also propose one concrete action — tagging those customers — and execute it on your confirmation (admins/owners only; the cohort is the result set you saw, capped at 200; proposals are derived server-side from the result shape, so they work identically on a keyless local model). Runs fully offline on a local model (Ollama or any OpenAI-compatible endpoint) — including the template-matching embeddings — or on your BYOK cloud key.
📉 Churn risk scoring	Per-item churn risk with suggested actions, plus cohort analytics and playbooks. Churn-triggered playbooks can now reach the customer: send_email playbook steps and bulk outreach campaigns on /customers (template or AI-drafted content, sent via your own Resend key), with a per-customer opt-out + tokenized unsubscribe honored on every send path and a shared per-recipient cooldown so no one gets emailed twice. An automation rule can send the same emails (send_customer_email) when churn risk, health score or usage crosses a line — shadow mode first, and every send is audited per rule. Optional CRM-sourced churn label suggestions — if you've connected HubSpot or Salesforce, Rereflect can read lost renewals and propose them as churn labels. Off by default and default-deny (nothing is suggested until you name your renewal pipelines), and every suggestion is confirmed by a human in a review queue before it becomes a label — a lost renewal is not always a churn. Includes an optional on-demand backfill over your closed-lost history. A second, CRM-free source does the same from sustained product-usage decline (opt-in, shadow mode first) — same review queue, same human-confirm rule; a usage drop is not always churn either.
🧑‍💼 Customer 360 + timeline	Per-customer health profile with a unified, paginated activity timeline (feedback, product usage, churn and health-score events); CRM enrichment (HubSpot, Salesforce) with optional health-score writeback to both HubSpot and Salesforce — also readable over the public API.
🗂️ Team workflow	Kanban board, statuses, auto-assignment rules and round-robin routing.
🔌 Sources & integrations	Inbound: CSV import, email, webhooks and Slack, plus support desks (Intercom, Zendesk). Issue trackers: Jira, Linear and Asana. CRM: HubSpot and Salesforce. Outbound: alerts and digests via Slack, Discord and Teams, and customer outreach via email (BYO-key Resend, opt-out honored).
📊 Trends over time & sharing	Not just a snapshot: /analytics plots average sentiment, volume, urgent count, pain points and feature requests over 7/30/90 days (daily or weekly buckets), and each top pain point and feature request carries an ↑/↓/→ arrow so you can see which themes are actually accelerating rather than just loud. Plus source breakdown, saved views, CSV export, PDF export and read-only share links.
👥 Multi-tenant + RBAC	Organization isolation with Owner / Admin / Member roles.

🚀 Quick Start

The fastest path is Docker Compose — it brings up Postgres, Redis, the backend, the Celery worker and the frontend together.

git clone https://github.com/haqaliz/rereflect.git
cd rereflect

# 1. Copy the env template and fill in the required secrets
cp .env.prod.example .env

# 2. Generate the two required secrets and paste them into .env
python -c "import secrets; print('JWT_SECRET=' + secrets.token_urlsafe(48))"
python -c "from cryptography.fernet import Fernet; print('LLM_ENCRYPTION_KEY=' + Fernet.generate_key().decode())"

# 3. Build and start everything
docker compose -f docker-compose.prod.yml up -d --build

Open http://localhost:3000 and sign in with the ADMIN_EMAIL / ADMIN_PASSWORD you set in .env (the first admin is seeded on startup). The API and interactive docs live at http://localhost:8000/docs.

Out of the box (ai_analysis_enabled=false, no LLM key) Rereflect runs the free local pipeline — sentiment, pain points, feature requests and heuristic churn all work with no external API and no cost. Point Settings → AI at a local model (Ollama or any OpenAI-compatible endpoint) and even the AI Copilot runs fully offline — its template-matching embeddings and answer generation both use your local provider, no OpenAI key required. Add a cloud BYOK key instead whenever you want a stronger hosted model.

👉 Full deployment guide, env reference and BYOK setup: docs/SELF_HOSTING.md. Developing locally instead of via Docker? See docs/DEVELOPMENT.md.

Screenshots

Dashboard & analytics

Real-time KPIs, NPS, sentiment distribution and trend analytics — with exportable, shareable views.

AI analysis: pain points, feature requests & urgent flags

Every feedback item is automatically categorized, tagged and prioritized.

Workflow board	Feedback sources
Integrations	AI (bring your own key)
Auto-assignment rules	Notifications & digests
Team & roles	Shared links

Architecture

┌─────────────────┐
│  frontend-web   │  Next.js 16 + TypeScript + TailwindCSS
└────────┬────────┘
         │ REST API
         ▼
┌─────────────────┐
│   backend-api   │  FastAPI + PostgreSQL + SQLAlchemy
└────────┬────────┘
         │
    ┌────┴────┐
    ▼         ▼
┌────────┐ ┌────────────┐
│analysis│ │  worker-   │  Celery + Redis
│-engine │ │  service   │
└────────┘ └────────────┘

A Next.js frontend talks to a FastAPI backend; long-running analysis runs on a Celery worker (Redis broker) using the analysis engine (VADER / scikit-learn / BERTopic, or an LLM when a key is configured). Full breakdown, tech stack, project layout and the RBAC model are in docs/ARCHITECTURE.md.

Documentation

Guide	What's inside
Self-Hosting	Docker Compose deployment, full env reference, BYOK, TLS and the $0 local mode
Development	Local setup, the pnpm + Python toolchain, package management, common commands, troubleshooting
API Reference	REST endpoints, auth, pagination and filtering (live Swagger at /docs)
Architecture	Services, tech stack, project structure and RBAC
Contributing	Dev workflow, testing and PR conventions

Tech stack

• Frontend — Next.js 16 · TypeScript 5.9 · TailwindCSS 3.4 · shadcn/ui · Recharts
• Backend — FastAPI 0.115 · SQLAlchemy 2.0 · Alembic · PostgreSQL · JWT
• Async — Celery 5.3 · Redis
• AI/ML — VADER · scikit-learn · BERTopic · OpenAI / Anthropic / Google (BYOK)

Contributing

Contributions are welcome — bug reports, features, docs and tests. See CONTRIBUTING.md for dev setup, testing and PR conventions.

Support the project

Rereflect is free and open source. If it's useful to you, you can chip in with a crypto donation:

License

Released under the MIT License. Third-party attributions are in NOTICE.