Demo polish + download-models.sh helper - #9
Merged
Conversation
Two real problems with the prior demo: 1. **Beat 3 asked about git stash inside a non-git sandbox.** Felt disconnected and the model wasn't sure how to be helpful — even if the answer came out clean, the framing didn't. 2. **Only two tiers ever appeared on screen** (fast as default, one `--balanced` for the piped diagnosis). The pitch is "you can pick fast, balanced, or smart per call" but the demo only ever showed two of three. This pass: - **Beat 0 is now `?? --status`** instead of `?? --start fast`. It prints all three tiers up with their ports (`fast → :8091`, `balanced → :8080`, `smart → :8093`), making the multi-tier capability visible from the first frame rather than implied later. - **Beat 3 question changed** to `? in 2 lines, what does the -k flag mean in du` — a flag the audience just saw in Beat 1's `du -k` output. Natural follow-through instead of a contrived "what does git stash do" in a directory with no git. - **Beat 4 routes to smart** (`? --smart`) instead of balanced. The Qwen3-Coder-Next tier is the right fit for the piped Python traceback, and using the smart tier here completes the fast/balanced/smart triad on screen (fast is implicit default, smart is the visible switch, balanced is in the status header). - **Archive seed is now a Docker volumes conversation** instead of git stash, and the recall (`??? docker volume`) finds it. The recall feature shows its actual value — pulling a real past conversation from days ago — without forcing a git context. No code changes. Just the .tape, the rendered .gif, and the seeded archive's topic.
shellllm's install used to say "huggingface-cli download
<long-repo-name>". That worked but it was a footgun: typing the
balanced repo wrong silently downloads nothing, and people couldn't
discover the fast/smart tiers without reading the docs.
`scripts/download-models.sh` wraps the same `huggingface-cli download`
call with the tier registry baked in. Behaviour:
scripts/download-models.sh # default: balanced
scripts/download-models.sh fast # one specific tier
scripts/download-models.sh fast balanced smart # multiple
scripts/download-models.sh all # every chat tier
scripts/download-models.sh all --with-embed # + the default embed tier
scripts/download-models.sh --with-embed=nomic # default chat + nomic embed
scripts/download-models.sh --list # what's downloaded vs not
The --list view shows every tier (chat + embed), the on-disk status
(✓/✗), the HF repo path, and which is the default. It's the same
information `?? --list` prints, but available before you've sourced
the zsh layer.
Plays nicely with cold installs: re-downloads are skipped because the
HF cache is content-addressed, so this script is safe to re-run.
Wired through:
- Homebrew formula installs it executable and points caveats at it.
- README install step 2 now calls the helper instead of the raw
huggingface-cli command.
- CONFIGURATION.md's "model tiers" section documents the helper.
bash 3.2 compatible (macOS default shell).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Two related changes that make the project more presentable to a first-time user.
1. Demo: drop git, contextual question, three-tier showcase
Problems with the prior demo
Changes
2. `scripts/download-models.sh` — one-command model fetch
The install used to say "huggingface-cli download <long-repo-name>". Easy to typo, hard to discover the other tiers from. The new helper wraps the same call with the tier registry baked in:
```sh
scripts/download-models.sh # default: balanced (~16 GB)
scripts/download-models.sh fast # one specific tier
scripts/download-models.sh fast balanced smart # multiple
scripts/download-models.sh all --with-embed # everything + default embed
scripts/download-models.sh --list # what's downloaded vs not
```
The `--list` view shows every tier (chat + embed), on-disk status (✓/✗), the HF repo path, and which is the default. Same information `?? --list` prints, but available before you've sourced the zsh layer. Re-downloads are skipped (HF cache is content-addressed), so the script is safe to re-run.
bash 3.2 compatible (macOS default shell).
Test plan