v0.2.5

Highlights

• Added OpenAI-compatible provider management for Tiarina API, OpenAI, OpenRouter, and custom providers.
• Added validated provider setup: providers are saved only after model fetching succeeds.
• Added safer provider management with API key refresh, model cache refresh, and custom provider removal.
• Added searchable model picker with provider headers, direct model selection, scoped provider search, and More Models.
• Added provider/model display in prompt footer and spinner status.
• Improved markdown streaming so responses no longer fragment mid-word or break markdown rendering.
• Improved diff previews with block-style add/remove rows and lightweight syntax highlighting.
• Improved terminal spacing for assistant answers, tool logs, validation output, and status summaries.
• Added <thought> / <think> handling with a compact Thinking Preview box while hiding raw tags from replies.
• Updated README and tests for provider/model management and rendering behavior.

Validation

• npm run build
• npm test
• npm run lint
• node dist/index.js --version

v0.2.0

v0.2.0 — Token Hardening, Stability, and Release Pipeline

This release focuses on reducing token waste, improving long-session stability, hardening file/tool behavior, and preparing the project for a cleaner release workflow.

Highlights

• Major token-efficiency improvements for tool results, file attachments, validation history, and session saves.
• Smarter and safer read_file behavior with caps, query excerpts, and binary/minified file guards.
• More robust OpenAI client behavior with timeout, retries, and clearer gateway error messages.
• Better session persistence with fallback titles and local compaction.
• CI, release scripts, expanded tests, and updated documentation.

Token & Context Efficiency

• Added stale tool-result eviction and summarization to avoid resending large tool outputs forever.
• Added per-tool-result cap with actionable hints for continuing via offset/limit or narrower queries.
• Added deduplication for repeated read_file calls on the same path/offset/limit.
• Added active cleanup for inline @file attachments after processing.
• Kept only the latest automatic validation failure in active/saved history.
• Added local serialization compaction for saved sessions.
• Added local /compact fallback that mutates active history when LLM compaction fails.
• Added max_tokens to main completion requests.
• Added input/output token tracking and cost estimate in /usage.

Tool Improvements

• Reduced read_file default cap to 500 lines / 30KB.
• Added read_file.query for targeted excerpts around matching lines.
• Added binary/non-text detection for read_file.
• Added minified/generated-file guard for default read_file reads.
• Added list_dir cap at 500 entries.
• Updated grep_search output to prefer relative paths.
• Added tests for read_file caps, query excerpts, binary/minified guards, list_dir, and grep_search.

Stability & UX

• Added streaming fallback flush for long paragraphs without blank lines.
• Added OpenAI client timeout and retry settings.
• Added clearer timeout/gateway error messages.
• Added synchronous fallback session title before background LLM title generation.
• Improved session save hardening with sanitized history.
• Added non-TTY CLI tests for --help and --version.

Docs & Release Workflow

• Added GitHub Actions CI for build, test, and lint.
• Added release:check and preversion scripts.
• Updated README/DOCS for session paths, tool behavior, Windows best-effort notes, and quality gates.
• Published package metadata now uses normalized bin path: aina -> bin/aina.js.

Validation

Validated before release with:

npm run release:check
npm pack --dry-run

Release check includes:

npm run build
npm test
npm run lint

### Published Package

- npm: ainacode@0.2.0
- CLI binary: aina
- Git tag: v0.2.0

### Notes

- Windows support is documented as best-effort until full native Windows TTY/ANSI testing is completed.
- Prompt caching remains deferred until gateway support for caching semantics is confirmed.
- Final-answer tool-schema suppression remains partial by design because the current loop does not add an extra final synthesis API call.

v0.1.0

First public release of ainacode (aina) — an AI coding assistant for your terminal.

Highlights

• Agentic tool-use: read/write/edit/multi-edit files, regex grep, glob find, streaming run_command, read-only git.
• Default / Plan / Auto modes (Tab to cycle) with safe confirmations, colored diffs, and dangerous-command guards.
• Streaming markdown rendering, automatic project-context loading (AINA.md/CLAUDE.md/AGENT.md), and post-edit auto-validation.
• Session persistence (/resume), undo (/undo), and slash commands incl. /compact, /usage, /status, /config, /commit, /diff.

Install
npm install -g ainacode
aina

Requires Node.js ≥ 18.