Skip to content

v0.16.0 docs

Choose a tag to compare

View all tags
@heinergiehl heinergiehl released this 17 Jun 10:25
v0.16.0
27e56e9

Release Notes: v0.16.0

v0.16.0 is a workflow-runtime and compound-request release. It adds structured multi-part request handling, makes pending workflow turns more authoritative, preserves schema-v2 structured form fields across frontend and backend compilers, and adds database guards around pending conversation state.

Highlights

  • Compound Request modes: bots can now choose legacy, shadow, or structured from the Admin UI. legacy remains the safe upgrade default, shadow audits structured plans, and structured executes registered capabilities.
  • Structured compound execution: action, Laravel AI tool, and API Connector-backed capabilities can be exposed as schema-driven read/write compound operations with bot-level allow-lists.
  • Safer AgentGraph usage: write, mixed, async, and threshold-sized structured plans can use durable AgentGraph execution, while small read-only plans stay synchronous.
  • Workflow turn ownership: pending workflow questions, pending interactions, compound approvals, continuations, interruptions, delays, and cancellations are resolved before session-memory recall or assistant fallback can answer.
  • AgentGraph interrupt repair: replacement turns now cancel the SDK run before starting a new run, and reconciler logic understands SDK interrupts without explicit interrupt IDs.
  • Schema-v2 form preservation: structured fields authored as JSON text in the semantic Ask inspector compile to collectForm nodes on both the frontend and backend compiler paths.
  • Database hardening: task frames now have source uniqueness, and MySQL/MariaDB installs receive generated-column one-pending guards matching the existing PostgreSQL/SQLite partial-index semantics.

What Changed

Compound Requests

  • Added explicit legacy, shadow, and structured engine modes.
  • Added shadow audit records for evaluating structured planning without changing visitor-facing behavior.
  • Switched structured planning to Laravel AI structured output with repair for single required string item inputs.
  • Added capability schema validation and safe execution for action, tool, and API Connector-backed plans.
  • Added planner relationship metadata for all, alternative, duplicate, dependent, and ambiguous plans.
  • Kept incomplete single-item plans on the normal workflow path so collect-input and human-in-the-loop nodes can pause and resume.
  • Added a bot-level engine selector and per-bot capability allow-list in Behavior > Compound Requests.

Workflow Turns and AgentGraph

  • Pending workflow ownership now runs before conversation recall in modern chat endpoints.
  • Interrupt replacements cancel AgentGraph-backed runs through the SDK before closing the local projection.
  • SDK interrupts without interrupt_id now reconcile against the same synthetic identity used by pending-interaction projection.
  • Compound confirmation and execution stay traceable through AgentGraph metadata on pending requests and interactions.

Schema-v2 Authoring

  • Ask steps with JSON-authored structuredFields now compile to collectForm runtime nodes.
  • Frontend and backend compilers preserve structured field names and slot contracts consistently.

Database Constraints

  • Added a unique source constraint for conversation task frames.
  • Added MySQL/MariaDB generated nullable guard columns for one pending interaction/request per conversation.
  • PostgreSQL and SQLite continue to use partial unique indexes for the same pending-state contracts.

Upgrade from v0.15.0

  1. Update the package constraint to ^0.16.0 or to the exact marketplace version you receive.
  2. Run composer update heiner/filament-agentic-chatbot --with-dependencies.
  3. Run php artisan migrate.
  4. Clear caches: php artisan config:clear && php artisan route:clear && php artisan view:clear.
  5. Run php artisan filament-agentic-chatbot:doctor.
  6. Merge published config changes for compound_requests.engine, compound_requests.graph.sync_item_threshold, and registered compound capabilities.
  7. Keep bots on Legacy fallback until structured capabilities have been verified in staging. Use Shadow audit before Structured execution for production bots.

Release Validation

The release gate for this line should include:

  • composer validate --strict
  • composer run-script pint:test
  • composer run-script stan
  • composer run-script test
  • npm --prefix resources/js/workflow-editor run test:contracts
  • npm --prefix resources/js/workflow-editor run build
  • composer run-script eval:compound-requests
  • composer run-script eval:workflow-turns
  • composer run-script eval:workflow-understanding
  • composer run-script release:marketplace-check

Fresh path-install, VCS tag-install, and host-app smoke tests should still be run before tagging and publishing the GitHub release.

Links

Assets 2
Loading