Version 8.0.2

Patch release: generated-code and agent-configuration fixes. It corrects SQL, backend, Django and agent output that failed when run, fixes agent configuration written by the editor, and links the three BESSER documentation sites to each other. There are no metamodel or API contract changes.

Generators

  • SQL for SQLite, MySQL, MSSQL and MariaDB. Models with an enumeration produced a PostgreSQL-only CREATE TYPE … AS ENUM statement for every dialect except Oracle, which SQLite and the others reject. The statement is now emitted for PostgreSQL only; other dialects keep the enumeration inline.

  • Action-language methods that change data. A method such as this.status = "read"; returned an error in the generated backend: the methods router did not import its own class’s update function, related links were sent as objects instead of ids, single-valued navigation returned a wrapper instead of the object, and a call on another object’s method named the wrong endpoint. These now work on standalone classes and on classes linked by associations. Generation no longer fails on a call to another class’s method or on a method that returns nothing.

  • Django on Windows. Generation stopped with a UnicodeEncodeError when its console output was redirected on Windows, and a failure inside generation was masked by a second encoding error. Console messages are now plain text and errors reach the caller.

Agents

  • Agents in a Docker Compose web app. A generated agent listened on localhost inside its container, so the editor’s chat could not connect through the published port. Agents bundled into the web-app image now bind to all interfaces; standalone agent generation is unchanged.

  • SQLite databases in the agent configuration. The editor wrote database: for an SQLite entry, while the agent framework reads file:. The editor now writes file:, and generation also accepts configurations saved by earlier versions.

  • User-profile conditions. A criterion such as age > 65 lost its comparator in the generated agent and became age = 65. Comparators are now kept, and generated code shows them literally.

  • Agent templates. The Gym Agent, Personalized Gym Agent and Chatbot templates used an LLM without defining one, so the generated agent failed at its first LLM call. Each now defines one.

Documentation

  • The BESSER, Web Modeling Editor and Modeling Agent documentation sites link to each other from every page, and broken links between them are fixed.

  • Production deployment notes that agent personalization needs a longer proxy read timeout (about 600 s): it makes one LLM call per selected option, so several options exceed the common 60 s default.

  • The editor documentation now states that Ctrl+Z does not undo changes made by the AI assistant.

Upgrade and deployment

No action is required. Operators who proxy the editor backend should give /besser_api/transform-agent-model-json a read timeout of about 600 s.