v0.14.9 - Claude Haiku 5.5 on Alysis Free

Claude Haiku 5.5 is now available in the Alysis Free model picker, replacing the retired Sonnet Free campaign entry.

• Native Anthropic Messages support, a 1M context window, and prompt caching.
• Adaptive thinking with low, medium, high, xhigh, max, and off. Auto uses medium.
• Existing account credits and saved model preferences are preserved.

Upgrade the terminal app:

pipx upgrade alysis-code

Then run alysis --version and choose Claude Haiku 5.5 in alysis config.

Full changelog

v0.14.8

What's Changed

• fix(release): prepare 0.14.8 with working production VSIX verification by @Perdikis10 in #32

Full Changelog: v0.14.7...v0.14.8

v0.14.7

What's Changed

• fix(release): prepare 0.14.7 with passing VSIX audit and responsive sidebar by @apfivos in #31

New Contributors

• @apfivos made their first contribution in #31

Full Changelog: v0.14.6...v0.14.7

v0.14.6

What's Changed

• Fix VSIX attestation and fresh-install runtime validation for 0.14.6 by @Perdikis10 in #30

Full Changelog: v0.14.5...v0.14.6

v0.14.3

Added

• Claude Sonnet 5.5 is available in the Alysis account model picker, with native Messages transport, streaming, signed reasoning, and tool continuation using the shared hosted credit allowance.

Changed

• Hosted accounts now default to automatic prompt caching. Legacy Manual settings migrate only when the other cache settings match the old defaults. Off, customized settings, and later explicit choices are preserved.
• Hosted Sonnet uses five-minute cache controls. Hosted DeepSeek, GLM, and Luna use provider-managed caching.

Fixed

• Cache previews and diagnostics follow the selected model's actual protocol.
• Hosted credit-limit errors retain their details, and Sonnet reasoning tokens are reported without double counting.
• HTTP MCP startup respects the configured timeout. Startup failures are reported once without retrying through classic chat.

Upgrade

uv tool upgrade alysis-code

Restart an existing TUI after upgrading so it loads the updated caching settings. See the hosted caching guide for migration details and controls.

Full changelog: v0.14.2...v0.14.3

v0.14.2

Fixed

• Hosted Luna now offers only supported reasoning efforts. Saved or environment-provided minimal and ultra settings map to low and max, preventing invalid-effort errors during the free-credit campaign.
• Hosted capacity and account request-limit messages retain the gateway's retry guidance.
• Fast mouse drags over working-tip links no longer open a link accidentally.

Upgrade

uv tool upgrade alysis-code

Full changelog: v0.14.1...v0.14.2

v0.14.1

Full Changelog: v0.14.0...v0.14.1

v0.14.0

Full Changelog: v0.13.8...v0.14.0

Alysis Code v0.13.8

Alysis Code 0.13.8 updates planning and persona controls and improves CLI reliability.

• Planning: legacy chat Plan Mode is removed. Use /persona architect for planning and Markdown documents, or /forge for a plan-driven execution workflow. Forge keeps its plan views and editor.
• Permissions and personas: /permissions replaces /mode. Permission selections take effect on the next message; persona changes wait for the active turn to finish. The footer shows pending permissions.
• Queued input: interruptions preserve undelivered messages and queued follow-ups. Shift+Left recalls the newest queued follow-up for editing.
• Reliability: preserve ordinary assistant replies in history, sanitize nested credentials in support bundles, paginate Git diffs safely, support longer background waits, and provide clearer verification and hosted-account errors.

Upgrade note: replace chat /mode commands with /permissions. Chat /plan commands are retired; /plan tasks|markdown|edit remains available inside Forge. The CLI --mode option is unchanged.

Changelog · Compare with v0.13.7

v0.13.7

Added

• Rotating usage tips above the chat input while the agent works, with clickable links.
  Tips hide during dialogs and stay out of conversation history.

Changed

• The context footer shows remaining conversation space after startup prompts and tools.
  Provider measurements and calibrated estimates inform the gauge, while capacity warnings
  continue to use the full request budget.
• Ctrl+J is the multiline input shortcut; Enter submits. The Alt+Enter alias is removed.
• Updated public documentation and contributor guidance.

Fixed

• Setup continuation hints stay visible above the footer on small terminals and long screens.
• Benchmark wheel builds preserve local build metadata and existing build artifacts, including
  when a build fails or is interrupted.