tankyu_sha

seeker in japaneesse

Status

Work is paused here and I really want to finish it off in next couple of months. Focusing on few different things at the moment.

What's this ?

1. tankyu-sha is is an AI assisted personal digest tool.
2. the end goal is to keep an eye over a bunch of stuff over the internet without having to go to the websites.

The overral system can be seen here plan.md, though it's not updated in a while. will fix that

How to run

Tools needed

• mise
• ollama
• have both llama3.2:3b plus nomic-embed-text models, but they can be changed via config.json
• chrome or bunx playwright install chromium

Running

• run mise install to get all deps
• mise run config:init
• mise run db:create
• mise run db:migrate
• run bunx playwright install to install chromium <- rn it tries to get system app, has to be configurable
• now we can run gleam run to start the main app
• run bun run dev to start the web app, eventually it'll be served as part of the main app
• visit http://localhost:3000/tasks

Roadmap (in no particular order)

• flatten all actor messages to separate actors <- rn all of them are blocked even though we can move ahead
• actor pool for all actors
• keep playwright instance alive and every request is a new conetxt + tab or reuse same context
• Kill playwright service after cooldown period
• finish search source
• expose all config options via UI It's exposed via priv/app_config.json now, use mise run config:init to generate it
  • summarry model, embedding model, or maybe a model per run
  • chromium location <- useful when running outside docker
• add courier actor to send digest
• some how add a setup to monitor system memory load and throttle LLM ops + browser ops
• add UI for home page task + source creation
• add UI for task list
• add UI for task_runs -> source_runs
• write a Dockerfile to distribute this