v0.1.0 — it installs, it runs, and it can tell you what it cost

v0.1.0 — it installs, it runs, and it can tell you what it cost

The first tagged release. A release number, not a capability milestone: M1 ("it works on my repo") is separately tracked in PLAN.md and is NOT met — nobody has yet watched Paider drive an end-to-end multi-file edit in someone else's repo.

Install

composer require paider/paider

What works

• paider chat — interactive session, six native tools behind an approval gate: read_file, write_file, patch_file, run_shell, git, artisan
• paider commit — stages, writes a message on the fast tier, commits
• paider cost — per-tier token and spend from an append-only event log, with the share column, the ratio line and the all-Opus comparison
• paider config:provider — switch between eleven verified model presets
• paider config:show — what am I actually running

The cost ledger is the point. Four named tiers — orchestrator, coder, research, fast — priced per model at write time and never re-priced on read, so replaying the log cannot rewrite what a past session cost. It reconciles against provider-reported usage; that is a live test, not a claim.

Verified against real providers: OpenRouter, Anthropic, and xAI's Anthropic-compatible endpoint.

What does not work / is not in this release

• curl|sh installer — planned as a standalone binary, not built yet
• FrankenPHP standalone binary — the binary has no embed step yet (a renamed FrankenPHP is FrankenPHP, not Paider), so it would have been a promise, not a download
• End-to-end edit in someone else's repo — never driven yet; this is where M1 ("it works on my repo") is tracked

Details

175 tests, 630 assertions, hermetic by default. Three more hit real providers on demand and cost real money, which is why they are opt-in.

Apache-2.0.