obsei
Open source, privacy-first Voice of Customer analytics. A self-hosted alternative to Enterpret and Unwrap that strips PII from feedback before any LLM sees it.
What is obsei?
Open source, privacy-first Voice of Customer analytics. A self-hosted alternative to Enterpret and Unwrap that strips PII from feedback before any LLM sees it.
Overview of obsei
What it is
obsei is an open source, self-hosted Voice of Customer analytics tool. It is a privacy-first alternative to Enterpret and Unwrap. It runs entirely inside your own infrastructure and strips PII from feedback before any LLM processes it. It supports bringing your own sources, models, and agents.
Key features
- Private by design: PII redaction, pseudonymisation, encrypted database, no telemetry, audit log.
- Bring your own sources, keys, and models (zero-shot classifier, decision model, or LLM via LiteLLM).
- AI-native: MCP server and Claude plugin for agents to answer questions with cited evidence.
- Multilingual support: feedback classified in its own language; multilingual model groups issues across 50+ languages.
- Wide range of sources and sinks: CSV, JSON Lines, REST, webhooks, App Store, Google Play, GitHub, Slack, Telegram, Discord, email, Jira, Linear, SQL databases, etc.
- Privacy controls: redaction of emails, phones, cards, IBANs, national IDs for 16 countries, optional name redaction, salted pseudonyms, encrypted DuckDB, forget/export/audit commands.
- Server capabilities: scheduled pipelines, webhook intake, MCP over HTTP, Studio, role-based access, SSO-proxy support.
Use cases
Analyzing customer feedback from reviews, tickets, calls, and surveys while keeping data private. Use cases include app review analysis, helpdesk support ticket triage, social listening, and air-gapped enterprise setups.
Getting started
bash pip install obsei[recommended] obsei init obsei try obsei run
License and pricing
License: Apache-2.0. Open source and free.
Based on the project's public README and repository metadata.
