Skip to content

Ygg v0.2.0-alpha

Pre-release
Pre-release

Choose a tag to compare

@achuthanmukundan00 achuthanmukundan00 released this 25 Jul 05:31
· 302 commits to main since this release

Ygg v0.2.0-alpha

Ygg v0.2.0-alpha is a local-first terminal coding agent release focused on
multimodal input, durable Responses sessions, native compaction, labeled custom
providers, and a calmer, more accountable terminal experience.

This is an alpha: configuration, session, extension, and public APIs may change
before 1.0. Ygg is not an operating-system sandbox. An enabled bash tool runs
with the authority of the user who launched Ygg.

Install

Ygg supports macOS and Linux. Install Rust 1.86 or newer and rg (ripgrep),
then run the pinned installer:

curl --proto '=https' --tlsv1.2 -LsSf \
  https://raw.githubusercontent.com/skaft-software/ygg/v0.2.0-alpha/scripts/install.sh | sh

For a manual install without changing a shell profile:

cargo install --locked \
  --git https://github.com/skaft-software/ygg \
  --tag v0.2.0-alpha \
  --bin ygg \
  ygg-coding-agent

Verify the installed version:

ygg --version
ygg --help

Highlights

  • Durable OpenAI Responses replay, authoritative raw-output persistence, and
    native compaction transport.
  • Multimodal image/audio reads with capability-aware ingestion and explicit PDF
    path handling for file tools.
  • Labeled custom OpenAI-compatible providers, including Apple Foundation Models
    support.
  • More focused reasoning and activity presentation with bounded, sanitized
    tool-output projections and edit/write diffs.
  • A corrected animated startup identity that reports the running package version,
    selected model, reasoning configuration, and workspace.
  • Terminal lifecycle, media, and transcript rendering fixes.

Compatibility and limitations

  • This release remains alpha software; configuration and public APIs may change.
  • Ygg sends model traffic directly from your machine to the endpoint you select;
    it does not provide a hosted control plane.
  • Command execution, when enabled, can read credentials, access the network, and
    start child processes with the launching user's authority. Use an isolated
    account, container, or VM for untrusted repositories or model endpoints.
  • Existing sessions containing historical exec calls remain renderable; new
    provider schemas use bash.

See the changelog
and security policy
for the complete changes and security boundary.