Course map
Fundamentals
You’ll understand the bot’s six layers (channels, queue, orchestrator, brain, Ollama, cost) and the exact path a message takes from Telegram to the result.
Install and operate
You’ll finish with v3 running as a systemd service, your own Telegram bot, a green doctor check, and the ability to read logs, the dashboard, and alerts when something breaks.
Use it every day
You’ll master the commands, queue, tasks, and cron, and understand how the brain stores memory, proposes facts for the vault, and ingests emails, calendars, and groups.
Configure and extend
You’ll know how to stop and prioritize the bot and give it a personality, measure the prompt’s fixed cost, create agents and skills, adjust the budget and Ollama, and understand what comes in phases 8 and 10.
The problem
V2 became a 145 KB monolith, with no cost tracking, no durable queue, and it ran out of memory twice in August. A personal assistant that goes down in the middle of the night isn’t an assistant.
The answer
V3 is being built alongside V2 to address the bottlenecks: a SQLite queue with leases and heartbeats, a single cost gateway, Ollama with a RAM preflight, and memory with nightly consolidation. Nothing is removed until parity is reached.
The result
A local Jarvis that replies on Telegram, delegates work to agents, remembers what matters, alerts you when something breaks, and shows how much it cost. All in about 8,000 lines of TypeScript with 187 tests.