Agentic Systems 2.0.0

Agentic Systems 2.0.0

Agentic Systems 2.0 delivers a coherent public grammar for tools, runtime Skills, agents, systems, environments, evals, providers, frameworks, CLI workflows, notebooks, and normalized RunResult contracts.

Release evidence

• GitHub Actions passes on Python 3.10 and Python 3.14.
• Full suite: 934 passed, 6 skipped, 100% statement coverage.
• Provider branch gate: 110 passed, 97.74% branch coverage.
• API contract: 78 public exports and the same checksum on Python 3.10/3.14.
• Release archives contain no .env, detected credentials, or legacy accountability_otc paths.

Assets

• Python wheel and source distribution for PyPI.
• Installable Agentic Systems skill ZIP for Codex/ChatGPT.
• Portable Agentic Systems Studio ZIP with component catalog and ten reference systems.
• SHA-256 checksum manifest.

Live OpenAI and Ollama evidence is included in the repository. Bedrock IAM/API-key and vLLM sandbox validation remain explicit deployment gates for ADA and Colab/GPU environments.

Agentic Systems 1.1.3

Agentic Systems 1.1.3 is the stable compatibility baseline before the 2.0 Provider × Framework architecture. It preserves the 111-symbol public API and all 1.1.2 public signatures while consolidating documentation, reorganizing tests by semantic owner, removing stale historical material, and making release checks version-independent. Verified with Ruff, 383 tests, 100% core statement coverage, 53.17% Bedrock coverage, 13 executed notebooks, Python 3.10/3.14 CI, twine check, and isolated wheel smoke.

Agentic Systems 1.1.2

Agentic Systems 1.1.2

Version: 1.1.2.

User Impact

This maintenance release makes Agentic Systems easier to trust and maintain
without changing how users call it. Existing code keeps the same 111-symbol
public API, the same agentic_systems.providers.bedrock_runtime import route and
the same public BedrockRuntime methods and signatures. No application
migration is required from 1.1.1-compatible code.

The package remains compatible with Python 3.10 through 3.14. Base imports stay
lazy with respect to boto3, OpenAI, LangGraph and other optional providers.

What Changed

The test suite no longer depends on dynamically loaded quarantine modules. All
52 routed modules were either migrated to their owning API/unit/release domain
or removed after their evidence became redundant. The final tree has no legacy
loader, quarantine directory, Ruff exception or numeric legacy guardrail.

The public providers/bedrock_runtime.py module is now a small compatibility
facade. Models, identity, tools, Converse/direct execution, OpenAI compatibility,
validation and LangGraph responsibilities live in focused internal modules under
providers/bedrock/. The facade remains the owner of the public class, and
compatibility tests freeze its callable signatures.

The detailed test mapping is recorded in
TEST_MIGRATION_1_1_2.md.

Release Evidence

• 381 tests pass with runtime warnings promoted to errors.
• 13 deterministic notebooks executed by pytest from clean kernels.
• 5 Provider notebooks checked statically for public-API, preflight and syntax contracts.
• Ruff passes over all src and tests without quarantine exclusions.
• Core coverage is 100% over 6,194 statements; Bedrock is explicitly outside
  that number.
• Separate Bedrock coverage is 53.17%, with a blocking fail_under = 53.1
  monotonic ratchet.
• The CI workflow gates Python 3.10 and 3.14 using
  .[dev,bedrock,langgraph,openai], without installing VLLM.
• Build, twine check and isolated Python 3.14 wheel smoke are release gates.

The Bedrock suite uses fake clients and requires neither live AWS credentials nor
network access.

Compatibility Guarantees

• agentic_systems.__all__ remains exactly 111 symbols.
• No runtime feature or public callable was added.
• Public Bedrock imports and signatures are frozen by contract.
• Fake-client injection through runtime.runtime and monkeypatching through the
  historical provider route continue to work.
• Wheel, CLI and optional imports are validated outside the repository.

Evidence Boundary

The automated suite proves deterministic behavior, packaging and adapter
contracts without claiming live OpenAI, Bedrock or vLLM execution. Provider
notebooks cross the external boundary only after an explicit readiness check;
an actionable skip is not reported as live evidence.