0.1.1

[0.1.1] - 2026-06-12

Fixed

• Inspector: add spacing between the model-call metrics grid and the cost section, so the
  cost details no longer appear visually glued to the metrics
  (#1).

0.1.0

0.1.0 - 2026-06-12

Initial release.

Added

• Logfire client (LogfireClient) — reads agent traces via the Logfire Query API
  (SQL + read token); region configurable with LOGFIRE_BASE_URL.
• Span-tree parser (parse_run) — turns the OpenTelemetry GenAI spans emitted by
  Pydantic AI into a typed WorkflowReport: conversation turns, model rounds, tool
  calls, and nested agents-as-tools (recursive, to any depth).
• Exact cost — per model call and per run, computed from tokens via
  genai-prices; reported as "unknown"
  when the model is not in the price database.
• Reasoning & token usage — per-call thinking summaries and input/output/reasoning
  token counts, aggregated for the whole run.
• Self-contained HTML report (render_html) — a Figma-style, pan/zoom/drag canvas
  with a single consistent Logfire-inspired palette, featuring:
  • block diagram of the full workflow with nested agent frames,
  • a guided tour with auto and manual (Space / click / arrows) stepping,
  • a click-through, resizable inspector and a full conversation transcript,
  • long content expanding into a modal; everything embedded, works offline.
• CLI — agentcanvas console script (and python -m agentcanvas): build the report
  from the latest or a specific trace, list recent runs, choose the output file.
• Example agent (assets/scripts/main.py) — a thinking agent with five tools, a
  nested sub-agent and a multi-turn conversation, plus scripts to record the demo media.
• Packaging, typing (mypy), linting (ruff), tests (pytest), and CI/CD (GitHub Actions
  with PyPI Trusted Publishing).