E

Release Notes — v0.2.19 (August 23, 2026)

1. New App: enezaagents — AI Agent Runtime

Status: Uncommitted

A brand-new Django app that lets users define, version, publish, and run AI agents from inside the Eneza admin, backed by pluggable LLM providers (OpenAI, Ollama) and tool-calling.

Models

Model Purpose
Agent Top-level agent definition with code, description, and status (Draft/Published/Archived). publish() snapshots the working draft into a new AgentVersion.
AgentVersion Versioned snapshot of an agent: instructions, model_config, output_schema, version_status, is_latest.
AgentVersionTool Through model binding a Tool to an AgentVersion with per-version configuration and enabled.
AgentVersionSubAgent Through model binding a child Agent as a delegatable sub-agent with a role and configuration.
LLMProvider Provider registry entry (OpenAI/Ollama/…) with JSON configuration and enabled.
ModelConfiguration Per-provider model settings: model_name, temperature, max_tokens, JSON configuration.
Tool Executable tool with a type (Python / REST API / Database), JSON schema, and configuration.
AgentRun Records one execution of a published AgentVersion, with input/output/status/task_id timestamps and optional parent_run for sub-agent nesting.
AgentStep Ordered trace of each step in a run (LLM call, tool call, sub-agent, complete, failed).
Artifact Named output artifacts produced by a run.

Provider Adapters (enezaagents/providers/)

  • BaseLLM ABC + LLMResponse/ToolCall normalization layer.
  • OpenAIProvider — chat completions, tool calling, structured output (json_schema), reads api_key/base_url from provider config or OPENAI_API_KEY/OPENAI_BASE_URL env vars.
  • OllamaProvider — ollama Python package, tool calling + structured output (format).
  • registry.get_llm() dispatches on LLMProvider.provider_type.

Runtime (enezaagents/runtime/)

  • agent_runner.run_agent_version() — iterative agent loop: builds tool/sub-agent schemas, calls the LLM, routes tool calls, records AgentRun/AgentStep, and terminates on final answer or max iterations.
  • ToolRouter — executes tools by type:
  • Python: imports a dotted callable path.
  • REST API: HTTP request via requests with params/data/json/headers config.
  • Database: safe queryset reads with field allowlists and an operator whitelist (exact, iexact, icontains, in, gt, gte, lt, lte, isnull), capped at max_rows.
  • Sub-agent delegation with cycle detection and max depth (MAX_SUB_AGENT_DEPTH = 5).

Tasks (enezaagents/tasks/run.py)

  • execute_agent_run background task (django-tasks) and queue_agent_run helper that defers enqueue until after commit and captures the current thread-local user.

Admin & URLs

  • Full CRUD setup views for all models via the framework setup/readonly base views, plus AgentPublishView (POST /agent/publish/<pk>/) and read-only list/detail views for AgentRun, AgentStep, and Artifact.
  • App menu registered in apps.py (Agents, LLM Providers, Model Configurations, Tools, Runs) with menu_position = 99.
  • Initial migration 0001_initial.
  • New template: enezaagents/templates/enezaagents/readonly_list.html.

Integration

  • enezaagents added to INSTALLED_APPS in framework/settings.py, eneza/templates/project/project_template/settings.py, and testproject/testproject/settings.py.
  • URLs mounted in framework/urls.py (/enezaagents/), project template (/agents/), and testproject/testproject/urls.py (/agents/), all with namespace='enezaagents'.
  • Packaging: enezaagents added to packages and package_data in setup.py.
  • New runtime dependencies: openai>=1.40.0 and ollama>=0.4.0 in setup.py install_requires and requirements.txt.

2. New BaseEnezaModelForm — Boolean Fields as Yes/No Select

Status: Uncommitted

Added BaseEnezaModelForm (enezaframework/forms/__init__.py) that automatically replaces every BooleanField widget with a new YesNoSelect widget (enezaframework/widgets.py) that renders a Yes/No dropdown instead of a checkbox.

All modelform_factory calls in UiMixin/UiMixinjournal (enezaframework/views/mixins.py) now pass form=BaseEnezaModelForm, so every generated model form inherits this behavior.

Display Consistency

get_smart_value (enezaframework/templatetags/custom_filters.py) now renders boolean values as Yes/No in read-only/detail views, matching the new form widget.

3. TomSelect Improvements

Status: Uncommitted

Auto-Initialization (enezaframework/static/assets/js/tomselect-init.js)

initTomSelects() is now invoked automatically: - on DOMContentLoaded when scripts load late, and - after every HTMX swap via the global htmx:afterSwap listener.

This guarantees the widget initializes regardless of load path and removes the need for per-partial inline init calls.

Detail Partials

Added if (typeof initTomSelects === 'function') { initTomSelects(); } to: - core/partials/detail_partial.html - core/partials/detail_source_document_partial.html

Styling (tomselect-custom.css)

  • .ms-wrapper now expands to fill available width (flex: 1 1 0%; min-width: 0; width: 100%).
  • Restyled control, search input, and dropdown colors to match the app theme (light #e0e6ed borders, dark #121e32/#17263c surfaces, primary accent #4361ee), larger padding, and heavier font weight.
  • Removed the dark placeholder override in favor of a single placeholder rule.
  • Cache-busting query param bumped to ?v=3 in components/layouts/default.html.

4. Sidebar Menu State Persistence

Status: Uncommitted

apps_sidebar.html now persists the active dropdown section per app:

  • Alpine $persist stores the open fasttab per app under eneza-sidebar-tab-<current_app> in localStorage.
  • On first visit (no saved choice), the sidebar auto-opens the section containing the active page.
  • Added data-tab attribute to menu buttons to support the auto-open logic.

5. Framework Test App (frameworkapp) Demo Changes

Status: Uncommitted

  • frameworktest gained a frameworktestsetup ManyToManyField (with related_name="frameworktest_setups").
  • Frameworktestsetup now has a __str__ returning its name.
  • FrameworkTestDetailView: frameworktestsetup added to the detail fieldtabs.
  • FrameworkTestCreateView: new tab group with the frameworktestsetup M2M field.
  • New FrameworktestsetupAutocompleteView (autocomplete for the M2M selector).
  • New migration 0011_frameworktest_frameworktestsetup.py.

6. Version Bump

  • setup.py: Version bumped from 0.2.18 → 0.2.19.
  • RELEASE_NOTE/CHANGELOG.md: Added 0.2.19 entry.

Summary

Area Description Files
enezaagents New AI Agent Runtime app (models, providers, runtime, tasks, views, URLs, menu, migration, templates) ~25
Framework Forms BaseEnezaModelForm + YesNoSelect widget, applied to all model forms 3
Boolean Display get_smart_value renders booleans as Yes/No 1
TomSelect Auto-init on load/HTMX swap, theme restyle, cache-bust 5
Sidebar Per-app dropdown state persistence via localStorage 1
Test App M2M frameworktestsetup, autocomplete view, migration 3
Integration INSTALLED_APPS/urls in 3 projects, packaging, deps 6
Versioning Version bumped to 0.2.19, CHANGELOG updated 2

Total: 20 modified files, 1 new migration, and the new enezaagents/ package (plus docs in .kilo/plans/ and feature_Planure/).

Note: All changes are uncommitted and untested. The user is solely responsible for testing and validating all outputs before committing or deploying.

View source Edit this page