Unarthi

Structured voice reflection for growth, stress, and inner clarity.

Sign in with Microsoft Entra ID
Tamil + English voice sessionsGPT-5.4 mini live repliesNeon + LanceDB memoryLiveKit Agents UI

A premium voice companion for becoming more honest with yourself.

Unarthi does not start every conversation from zero. It keeps a structured, evolving psychological profile shaped by real frameworks, careful evidence, and gentle voice reflection instead of generic motivational chatter.

Persistent model of you

Profiles values, distortions, attachment patterns, regulation, and change readiness.

Human-designed psychology layer

Frameworks are programmable, but the interpretation stays bounded and reviewable.

Clear safety boundary

Built for reflection and growth, not for pretending to replace therapy or crisis care.

Real-time Voice Companion

Unarthi listens for patterns, not just prompts.

The moat is the structured profile: frameworks for personality, cognition, values, regulation, stress, and growth, updated session by session and kept grounded in human review.

Live LLMgpt-5.4-mini
VoiceLiveKit + ElevenLabs
StorageNeon + LanceDB

Session Bootstrap

Choose your session language

The current voice path is stable in both English and Tamil. Pick the language you want Unarthi to stay in for this session.

Deployment note

Core frameworks can actively shape session guidance and weekly summaries in v1.

Session language: English.
Sign in with Entra ID to start.

Profile Snapshot

Unarthi is still building a structured read on recurring patterns and values.

Top values

No values pinned yet.

Recurring distortions

No strong distortion pattern yet.

Safety: stableLast profile write: Just nowFrameworks active: 0

Session History

Start a session and Unarthi will keep the summary trail here instead of losing it with the room.

Memory Ledger

No memory snippets stored yet. Completed sessions will start to accumulate reusable summaries here.

Signal Trail

Core frameworks shape the response engine immediately. Extended and experimental layers stay visible but should remain human-reviewed until you validate them with more sessions.

Live conversations use gpt-5.4-mini; deeper profile analysis runs on gpt-5.4-pro. Data is stored in Neon + LanceDB.