Datalk

Talk with your data!

Available flake outputs

• nix run .#tf-[apply|destroy]: deploy/destroy terranix configuration
• nix run .#containers -- create --start: spin up docker compose-like development containers (eg for db, redis)
• nix run .#ui: build & launch the SvelteKit application
• nix run .#python-server: build & launch the remote-execution python server
  • NB: This is explicitly remote code execution over http! Must sandbox!
• nix run .#push-images-local: build & push the production-style & development local k3d images
• nixosConfigurations.ui: main NixOS system serving the SvelteKit application
• nix develop:
  • Install pre-commit-hooks
  • Install LSP / package build inputs
• nix fmt:
  • TS/Svelte support pending!

Available datasets

The python/sql execution environment expects a "dataset connector" to be a folder of .csv files. Each .csv will be loaded into a "table" / dataframe, and all tables within a dataset will be made available to any chat with that datasetenabled (1 chat <=> 1 dataset).

Currently, only a subset of the excellent (College Football Data)[https://collegefootballdata.com] (2025 only) has been "connected." More datasets could be easily uploaded to the sandbox container -- but I am concerned about overloading my limited computational resources in production!

Local UI hot reload on k3d

The local Kubernetes environment can run the UI from a Nix-built dev image when hotReload = true in nix/kubernetes/default.nix. It mounts this checkout into the k3d nodes at /workspace/datalk. The pod mounts ui/src, ui/static, and ui/drizzle from there, so SvelteKit/Vite can reload without rebuilding the image for normal source edits.

If the k3d cluster already exists from before this mount was added, recreate it once with nix run .#tf-destroy-local and nix run .#tf-apply-local.

Use nix run .#push-images-local when Node dependencies or UI config baked into the dev image change.