Changelog

All notable changes to TyCode are documented here.

[0.2.0] — 2026-05-30

GitHub release: v0.2.0

Interruptible agent: Esc (or Ctrl+C) now aborts the in-flight model
stream mid-token and stops the agent loop immediately. Partially streamed text
is finalized and remaining tool calls get synthetic results so the transcript
stays valid for the next turn.
Automatic retry with backoff: transient provider failures (HTTP 429/5xx,
rate limits, network/connection/timeout errors) retry up to 4 times with
exponential backoff and jitter instead of killing the run. Each retry is shown
inline and is itself cancellable.
Concurrent tools: independent read-only tool calls (file_read, grep,
glob_search, directory_tree, web_fetch, …) in a single batch run in parallel —
a multi-read fan-out now completes in roughly the time of the slowest call.
Mutating tools and dangerous shell commands stay sequential and gated on
confirmation, preserving correctness and ordering.
Streaming efficiency: SSE/NDJSON parsing in the Anthropic, OpenAI, and
Ollama providers no longer reallocates the whole buffer per event
(O(n²) → O(n) via drain); the OpenAI path also no longer risks
double-parsing a record split across chunk boundaries.

todo_write / todo_read: structured, session-persistent task list.
multi_edit: multiple exact-string edits to one file, applied atomically.
web_fetch: fetch a URL and strip HTML to readable text.
Project memory now auto-loads CLAUDE.md and AGENTS.md alongside
TYCODE.md / README.md.

Ctrl+Enter tool-call expand/collapse was unreachable (shadowed by plain
Enter) — now works.
Cleared all outstanding compiler warnings; the build is warning-free.

Initial public release.

Multi-provider agent: Ollama (local), Anthropic, OpenAI, Google Gemini.
Tool suite: file read/write/edit, shell with real timeouts, grep, glob,
HTTP, process management, system info, directory tree.
Autonomous agent loop with streaming responses.
TUI with markdown rendering, inline expand/collapse of tool output,
mouse scroll, dangerous-command confirmation, and auto context compaction.