Skip to content
Riverse

First Chat

After installation and configuration, here's how to start your first conversation with Riverse.

CLI Mode

The simplest way to get started:

source .venv/bin/activate
python -m agent.main

Type your messages directly in the terminal. Type quit or press Ctrl+C to exit — Sleep will automatically run to consolidate the conversation into memory.

Telegram Bot

python -m agent.telegram_bot

Open your Telegram bot and start chatting. Supports:

  • Text messages
  • Voice messages (transcribed via Whisper)
  • Images (analyzed via GPT-4 Vision / LLaVA)

Send /new to start a new session and trigger Sleep.

Discord Bot

python -m agent.discord_bot

REST API

uvicorn agent.api:app --host 127.0.0.1 --port 8400

Send a message:

curl -X POST http://localhost:8400/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello, nice to meet you!"}'

Web Dashboard

python scripts/start_local.py          # Recommended: starts FastAPI + Flask together

Or manually (both required):

uvicorn agent.api:app --host 127.0.0.1 --port 8400
python web.py

Visit http://localhost:1234 to view:

  • Profile overview (categories, timeline, confirmed/pending status)
  • Relationship graph
  • Trajectory analysis
  • Observations
  • Memory review (confirm, reject, resolve contradictions)
  • Finance and health data

Tips

Recommended: Telegram Bot

Since Riverse handles text, voice, and images, Telegram Bot is the recommended primary interface. Set your unique Telegram User ID in settings.yaml.

On accuracy

No LLM today is specifically trained for personal profile extraction, so results may occasionally be off. When you spot something inaccurate, you can reject incorrect memories or close outdated ones in the Web Dashboard. Riverse intentionally does not allow manual editing — wrong memories are like sediment in a river, meant to be washed away by the current, not sculpted by hand. As conversations accumulate, the River Algorithm continuously self-corrects.