Skip to content

Releases: rrahimi-uci/agentic-context-engineering

Release list

v0.3.0

v0.3.0 Latest
Latest

Choose a tag to compare

View all tags
@rrahimi-uci rrahimi-uci released this 29 Jun 20:55
v0.3.0
bb02821

ACE v0.3.0

Backward-compatible release hardening the library and its tooling.

Reliability & DX

  • OpenAILLM hardening — native JSON mode (response_format), built-in retries/backoff, and a request timeout.
  • Non-blocking async — arun_and_learn / arun_streamed_and_learn run the Reflector/Curator off the event loop (asyncio.to_thread).
  • Parallel evaluate(max_workers=…) — concurrent, order-preserving, result-identical inference pass.
  • Auto-wired semantic embedder for grow-and-refine dedup when an OpenAILLM backend is present (config auto_embedder; lexical fallback).

Observability

  • Token/cost surfacing — StepRecord & RunResult.summary() report llm_calls, prompt/completion tokens, and cached_prompt_tokens (OpenAI's automatic prefix caching of the static system + playbook prefix).

Project quality

  • CI quality gates — ruff (lint + format), mypy (backing py.typed), and coverage (~95%, ≥90% floor).
  • All GitHub Actions on Node 24; version single-sourced from pyproject.toml; added CHANGELOG.md.

161 tests, fully offline by default.

Full diff: v0.2.0...v0.3.0

Assets 2
Loading

v0.2.0

v0.2.0

Choose a tag to compare

View all tags
@rrahimi-uci rrahimi-uci released this 29 Jun 17:12
v0.2.0
2561cef

ACE v0.2.0

Backward-compatible feature release.

OpenAI Agents SDK integration

  • wrap_agent() — one-call entry point: builds the ACE engine, injects the playbook, and persists what it learns.
  • Auto-learn from tool errors via RunHooks, rich trajectory capture using the SDK's typed run-items, an ace.learn tracing span, streaming, async, and session passthrough.
  • Top-level lazy re-exports (from ace import wrap_agent, ACEAgent) — import ace stays SDK-free.
  • openai-agents pinned to a realistic floor (>=0.1,<1.0).

Cookbook

  • 10 guided, runnable recipes (7 offline + 3 agent), each tested.

CI / tooling

  • Allure test reporting published as a CI artifact.
  • All GitHub Actions bumped to Node 24 majors.

148 tests, fully offline by default.

Full diff: v0.1.0...v0.2.0

Loading

v0.1.0

v0.1.0

Choose a tag to compare

View all tags
@rrahimi-uci rrahimi-uci released this 27 Jun 18:13
v0.1.0
3620711

First public release of ace-playbook — Agentic Context Engineering as a library.

Highlights

  • OpenAI Agents SDK integration (ace.integrations.openai_agents.ACEAgent): drop-in self-improving memory with sync and async entry points (run/arun, run_and_learn/arun_and_learn), string or dynamic base instructions, and label-free execution feedback.
  • Core ACE loop: Generator → Reflector → Curator with deterministic delta merge and grow-and-refine (no context collapse).
  • Fully offline, deterministic demos and 112 tests (zero network, no API key).
  • ace CLI: demo / run / playbook / version.

Install

pip install ace-playbook            # core (numpy + rich)
pip install "ace-playbook[all]"     # + openai + openai-agents
Loading