• Added notification preferences SDK methods:
  • update_notifications(...) / update_notifications_async(...) -> POST /v1/me/notifications
  • supports partial updates for email_notifications_enabled, tool_error_notifications_enabled, run_failure_notifications_enabled
• Added tests for notification preferences client methods (sync + async payload validation and endpoint mapping).
• Updated README.md and examples/README.md for latest backend plan matrix:
  • tools registry now available on all plans with limits 1 / 20 / 100 / 1000
  • governance entitlements and kill switch availability aligned with backend
  • documented free hard-limit vs paid soft-limit usage behavior

onceonly-sdk v3.0.4 (2026-03-25)

Added

• post_event(...) / post_event_async(...) for POST /v1/events
• get_run_timeline(...) / get_run_timeline_async(...) for GET /v1/runs/{run_id}
• High-level wrappers: ai_run(...) / ai_run_async(...)
• Optional run_id in AI high-level calls:
  • ai.run*, ai.run_and_wait*, ai.run_tool*
  • Auto-propagation to metadata.run_id (key mode) and args.run_id (tool mode)

Improved

• events(...) / events_async(...) now support offset
• New debug examples:
  • examples/ai/run_debug_timeline.py
  • examples/ai/run_debug_failure.py
• README and examples docs updated for run timeline/debug workflows

Compatibility

• Backward-compatible release, no breaking changes.

onceonly-sdk v3.0.3 (2026-03-10)

Changed

• Updated repository links after migration to GitHub org OnceOnly-Tech/onceonly-python.
• Updated package metadata repository URL used by PyPI project links.

Notes

• No API or behavior changes.
• This release is focused on repository/org migration metadata.

• Documentation updates (README links and references)
• Minor client compatibility improvements (base_url normalization, clearer 403 feature gating)

• Updated README links and API reference

Highlights

• Added tool registry support (create/list/get/toggle/delete) in SDK
• Added governed tool runner + full LLM agent flow examples
• Added UX+AI docs: cheat‑sheet, result shapes, plan limits, Pro vs Agency
• Added release checks + integration smoke tests

Docs & Examples

• Full LLM flow examples: without OnceOnly vs with OnceOnly
• Plan limits, policy templates, tools registry usage

Quality

• New tests for tools registry and integration smoke checks

Docs & examples

• Updated README with AI-first guidance
• Expanded README for examples/ai/

Breaking changes

None (backward compatible)

This release finalizes the AI API and stabilizes SDK behavior for agents and long-running jobs.

Highlights

• Finalized AI client with high-level run_and_wait() and low-level AI Lease API
• Exactly-once local side-effects via AI leases (payments, emails, webhooks)
• Canonical AI examples (examples/ai/) covering jobs, polling, results, and LangChain tools
• Clear separation between automation (check_lock) and AI usage (AI plans)
• Improved documentation and README aligned with AI API v2.x
• Comprehensive tests for parsing, fail-open behavior, and LangChain integrations

Notes

• AI usage is charged once per key; polling is free
• Fail-open applies only to network/5xx errors (not auth, limits, validation, or rate limits)

Recommended for all users building AI agents or distributed workflows.

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog
and this project adheres to Semantic Versioning.

[v2.0.0] — 2026-01-XX

Added

• AI-first SDK architecture
• Stable idempotency hashing for:
  • Pydantic v1 / v2 models
  • Dataclasses
  • Bytes / complex structured inputs
• Optional LangChain integration:
  • make_idempotent_tool
  • Support for StructuredTool (multi-argument tools)
• AI helpers for long-running backend jobs:
  • ai.run_and_wait
  • ai.run_and_wait_async
• Automatic Retry-After parsing for 429 responses
• Configurable retry logic for rate limits
• Metadata support for tracing (agent id, trace_id, custom context)
• Clear example separation:
  • examples/general/
  • examples/ai/

Changed

• Complete internal refactor of client architecture
• Unified error handling and response parsing
• Improved fail-open logic for production safety
• More explicit and descriptive docstrings (AI / agent friendly)
• Examples rewritten to reflect real-world AI-agent usage

Fixed

• Unstable hashing caused by repr() memory addresses
• LangChain tool execution issues with multiple arguments
• Retry-After header handling inconsistencies
• Edge cases in duplicate detection under retries

Removed

• Legacy v1 SDK stru

New

1. README update