Skip to content

Releases

v0.2.1

v0.2.1 Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 30 Oct 13:09
v0.2.1
e41f1c9

A small fix release.

Fixes

  • htd probe checks your configured model by default, not port 8000.
  • htd replay and htd bench summaries say what each server error was.
  • The between-jobs server check no longer fills your model server's log with errors.

Install

curl -fsSL "https://github.com/HTMLGuyLLC/HometownDiscount/releases/latest/download/htd_$(uname -s | tr A-Z a-z)_$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/').tar.gz" | tar -xz htd
sudo mv htd /usr/local/bin/
htd setup

Check a download with gh attestation verify <file>.tar.gz --repo HTMLGuyLLC/HometownDiscount, or compare it with SHA256SUMS.

Full changelog: v0.2.0...v0.2.1

Assets 7
Loading

v0.2.0

v0.2.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 28 Oct 02:55
v0.2.0
aff6d7a

An early preview of Hometown Discount: find out how much of your Claude Code work your local models can actually do, proven on your own jobs and your own code. Works end to end on macOS and Linux.

New

  • Several local models side by side. Set up a big model next to a small, fast one, or a few that each do some things well. Each model earns its own proof per task and size, and ask_local hands each task to the fastest one that has proven it and is within your speed limit. [pick] in config.toml lets you prefer the model listed first, or pin a task to a model. htd setup offers to add more.
  • Thinking models, checked both ways. A model that thinks before it answers is also checked with thinking off, which is often far quicker, and ask_local uses whichever way is proven and faster for each skill and size.
  • Speed counts, not just accuracy. A proven task goes local only if the model is within your limit of Claude's own time on that kind and size of task (by default, at most 20% slower), measured from your own transcripts. Set limits per task and size under [speed].
  • Results are tied to the Claude model a replay is graded against, so matching one Claude model doesn't count as matching another.
  • Snapshots (htd hooks on, offered by htd setup): your repo's uncommitted files are saved as each subagent starts, so replays can grade jobs that read them.
  • htd card shows what ask_local has saved so far.
  • Low memory pauses a run. htd bench and htd replay wait for memory to recover instead of pushing your machine, then stop with finished work kept.

Security

  • Replayed commands get a process-count limit, and nothing they start outlives them.
  • Local host names are checked again each time a connection opens.
  • Fixes from an independent security review. See SECURITY.md.

Changes

  • htd bench and htd replay run every model in config.toml; --model or --url runs just one.
  • Add ask_local to every project with claude mcp add --scope user htd -- htd mcp.
  • A misspelled setting in config.toml now stops htd with its name. Settings older versions wrote and nothing used are dropped the next time the file is saved.

Install

curl -fsSL "https://github.com/HTMLGuyLLC/HometownDiscount/releases/latest/download/htd_$(uname -s | tr A-Z a-z)_$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/').tar.gz" | tar -xz htd
sudo mv htd /usr/local/bin/
htd setup

Check a download with gh attestation verify <file>.tar.gz --repo HTMLGuyLLC/HometownDiscount, or compare it with SHA256SUMS.

Full changelog: v0.1.0...v0.2.0

Loading