Skip to content

Docyrus CLI v0.0.79

Choose a tag to compare

@anilbeyazoglu anilbeyazoglu released this 10 May 22:13

Docyrus CLI v0.0.79

Published to npm as @docyrus/docyrus@0.0.79.

npm install -g @docyrus/docyrus@0.0.79

Docyrus CLI v0.0.79

A big upgrade for docyrus server: pi runtime jumps to v0.74, the model
picker honours each model's real thinking-level capability, and logging out
of a provider actually sticks.

✨ New

  • Per-model thinking levels in /api/models. Each model now reports the
    exact thinking ladder it supports (off → minimal → low → medium → high → xhigh) so chat clients can offer just the levels that work, instead of
    showing a fixed list and silently clamping at the provider.
  • OpenAI Codex models exposed in server mode. Logging in with
    openai-codex (ChatGPT Plus/Pro Codex Subscription) now surfaces
    gpt-5.5, gpt-5.4, gpt-5.4-mini, and gpt-5.3-codex in
    /api/models. Older gpt-5.1/gpt-5.2 variants are intentionally
    filtered out.
  • All your configured providers are visible. /api/auth/providers
    now lists every provider pi knows about (Anthropic, OpenAI, Azure,
    DeepSeek, MiniMax, Moonshot, GitHub Copilot, custom OpenAI-compatible,
    Amazon Bedrock, …) so the UI can render the full login catalog instead
    of a curated subset.

🚀 Improvements

  • Pi runtime upgraded to v0.74.0. Brings JSONC-style models.json
    parsing (comments + trailing commas), incremental bash output streaming
    in the TUI, interactive OAuth login choices, post-turn stop callbacks
    for extensions, and a long tail of provider/network fixes from upstream.
  • Custom DeepSeek thinking levels survive the upgrade. If your
    ~/.docyrus/pi/agent/models.json still uses the old
    compat.reasoningEffortMap, the CLI rewrites it on first launch into
    the new model-level thinkingLevelMap automatically — no manual edit
    required, and your per-level mappings (including xhigh) are
    preserved.

🐛 Fixes

  • Provider logout now actually logs you out. Calling
    POST /api/auth/providers/:id/logout previously left the provider's
    API key sitting in the live process and in models.json, so the
    provider kept showing configured: true until you restarted.
    Logout now clears auth.json, env.json, process.env, and the
    models.json provider block in one shot, and the running server's
    model registry refreshes immediately.
  • Stale provider state is cleaned up at startup. If auth.json was
    edited out-of-band (or pi-mono's own /logout ran without the
    Docyrus wrappers), any leftover env vars and models.json provider
    blocks for managed providers (deepseek, azure-openai-responses,
    custom-openai, amazon-bedrock) are pruned automatically when the
    CLI or server starts.
  • docyrus server no longer leaks the global Docyrus auth into
    project-local sessions.
    Endpoints like /api/env/status and
    /api/souls now use the same scope (local vs --global) the CLI
    is running under, instead of always reading from ~/.docyrus/.
Technical changes

✨ Features

  • pi 0.74 + per-model thinking levels and sticky logout, bump 0.0.79 (e212ea316)
  • webform_record table and types generated for tenant schemas (ab587da14)

🐛 Bug Fixes

  • suppress pi TUI skill collisions from .agents/skills/ (bcadf40b7)
  • pass settings root explicitly to agent server (66e79d717)

Full changelog: Docyrus/docyrus-core@cli-v0.0.78...cli-v0.0.79