Next step is giving it temporal memory - it needs to understand time to learn patterns. Multiple memory systems with different roles at the same time: session (current short term memory of the session with the user - the current day or hours), goals (what has the user said he wants to do / achieve), movement (things that have changed in meaningful ways, problems, success, what are we doing to achieve our goals), patterns (things which might be useful to save to find patterns later like requests etc if they happen in the morning next time the user might not have to say anything the agent does it automatically), actions (things the user has done, things i have done)
all memories are chronological.
cd orchestrator bun run index.ts
t2: cd ui bun run dev
t3: cd telegram bun run index.ts
see .ts .svelte .js .toml see .ts .svelte .js .rs .toml -e airplane -e radar-controller -e lidar-controller -e ai-ui -e ai-ui-ui
see .ts -e airplane -e radar-controller -e lidar-controller -e ai-ui -e ai-ui-ui -e protcols -e tts -e tts-ui -e ui -e telegram -e telemetry -e verify-types-tests -e memory -e kindly -e discord -e architect -e assets -e coder
see .ts .rs .toml -e airplane -e radar-controller -e lidar-controller -e ai-ui -e ai-ui-ui -e protcols -e tts -e tts-ui -e ui -e telegram -e telemetry -e verify-types-tests -e memory -e kindly -e discord -e architect -e assets -e coder
When you add a Cell.toml the orchestrator picks it up and runs the spawn script you write in it. Prototyping meshes is very fast with this... But it can be faster.
Can this ever be safe? Like if I create a auth cell which is the load balancer and api entry point and only exposed cell which only routes calls to the rest of the mesh only if its authenticated?