mimi

LLM-driven multi-source assistant for Cyber Valley.

home: cyberia/research/mimi · remote: cyberia-to/mimi

Mimi is an LLM-driven, multi-source assistant designed to unify and enrich knowledge aggregation for Cyber Valley and similar dynamic projects. It automates retrieval-augmented generation (RAG) and enables natural language interaction with diverse sources such as Telegram, Logseq, and GitHub.

Services

• LLM-powered Chatbot
  Telegram agent with context-aware chat, topic following, metadata filtering, and periodic summarization. Model-agnostic: uses OpenRouter or Gemini (via Genkit).
• Scrapers & Data Synchronization
  • Logseq: Git repo-based knowledge base, parsed & indexed to CozoDB.
  • GitHub: Watches events/issues/boards and synchronizes project data.
  • Telegram: Ingests group/forum messages using Telegram Client API.
• RAG Engine
  Stores and retrieves relevant content from all sources; operates over structured and text data.
• Summarization Agents
  Generate daily, weekly, and topic-based reports.

Technologies & Architecture

• Codebase: Go 1.24
• LLM Integrations: OpenRouter, OpenAI, Gemini, via Genkit
• Vector Search & DB: Hybrid
  • Metadata, settings, chat: PostgreSQL (via pgx)
  • Graph/semantic queries: CozoDB
• Containerization: Podman/Docker/OCI
• Configuration: .env / example.env, mimi_config.json
• Infrastructure-as-Code: Ansible (services, Postgres container, config)

Key Design Decisions

• Model/Provider Agnostic: Easily swap LLMs or embedding providers via config/environment.
• Source-Agnostic RAG: All documents from Telegram/Logseq/GitHub are stored raw for future (re-)embedding.
• Metadata Enrichment: Structured metadata/tagging at ingest for granular filtering.
• Full Conversation Memory: Chat state/history stored as JSONB in Postgres; selective context windowing.
• Logseq Native Graph: Implements own Logseq parser/graph sync and custom Datalog-like querying.
• Full Infrastructure Automation: Use Makefile, Ansible for build/run/deploy/migrate.

Usage

Development

• Build:
  make install, make dev-db, make migrate-up make run (start bot & scrapers)
• Format/Lint/Test:
  make format, make vet, make test
• Migrations:
  make migrate-up / make migrate-down (Geni, SQLC, etc.)
• Deployment: make -C ansible/ deploy-service

Deployment

• Container:
  podman build -t mimi .
• Env/config:
  Configure all relevant API keys and DB params in .env/example.env
• Ansible:
  • ansible-playbook ansible/postgres.yml (start DB)
  • ansible-playbook ansible/server.yml (install deps)
  • ansible-playbook ansible/service.yml (build/deploy service)

Source Overview

• cmd/app/ — main entrypoint (Telegram bot, orchestration)
• cmd/scraper/{github,logseq,telegram}/ — resource-specific sync services (mostly for the testing)
• prompts/ — system/user prompts for RAG and LLMs
• internal/bot/ — bot logic, context, LLM/pluggable agents
• internal/provider/{github,logseq,telegram}/ — data adapters, scraping, parsing
• internal/persist/ — Auto generates sqlc queries from sql/queries
• ansible/ — automation for DB, network, service deployment

Maintained Technologies

• Go, Genkit, OpenRouter, OpenAI, Gemini
• PostgreSQL, CozoDB
• Podman, Ansible, Make, SQLC, Geni
• Telegram Client/HTTP APIs, GitHub APIs, Logseq

Example .env variables

See example.env

See Also

• Makefile
• Containerfile
• ansible/ — Playbooks for infra

Status: Alpha · Multi-source RAG platform