LightAgent v0.9.0

LightAgent v0.9.0

LightAgent v0.9.0 upgrades LightFlow from in-memory orchestration toward persistent, recoverable workflows, while strengthening memory admission controls and adding reusable guardrail templates for safer production deployments.

Highlights

• Add persistent LightFlow checkpoints with JsonLightFlowStore.
• Add workflow recovery APIs: flow.resume(run_id), flow.rerun_step(run_id, step_name), flow.get_run(run_id), and flow.list_runs().
• Expand LightFlow step state tracking with pending, running, success, failed, skipped, and waiting_approval.
• Improve DAG validation for unknown dependencies, cycles, and isolated steps.
• Add step-level controls for timeout, cancel_if, fallback_agent, requires_approval, and approval_handler.
• Enrich flow traces with step input/output summaries, retry count, error reason, duration, fallback usage, and run IDs.
• Add SharedMemoryPool and SharedMemoryRecord as a lightweight shared memory prototype for multi-agent experiments.
• Extend MemoryPolicy with expiration-aware retrieval, minimum write length, duplicate-write rejection, and reject-pattern controls.
• Add reusable Guardrails templates for privacy-sensitive input blocking, sensitive tool confirmation, high-risk parameter validation, and output redaction.
• Update LightFlow, memory admission, shared memory, guardrails, roadmap, and README documentation.

Compatibility

• Existing agent.run() string-return behavior remains unchanged.
• Existing agent.run(..., stream=True) behavior remains unchanged.
• Existing memory backends using store(data, user_id) and retrieve(query, user_id) remain supported.
• Existing LightFlow().step(...).run(...) workflows remain compatible; result objects now include richer status and trace metadata.
• New stricter memory and guardrail behavior is opt-in through MemoryPolicy and guardrail configuration.

Tests

• GitHub CI passed on Python 3.10, 3.11, and 3.12.
• Local targeted regression coverage included v0.6.5 core behavior, v0.7.0 tracing, memory policy, shared memory, guardrails, LightFlow, LiteLLM client, skill manager logging, and vector memory adapter examples.