Cerb (11.1) is a feature upgrade in development as of April 02, 2025. It includes more than 34 new features and improvements from community feedback.

• Important Release Notes
  • Added
  • Changed

Important Release Notes

• Always make a backup before upgrading!

• Cerb 11.1 requires PHP 8.2+ and MySQL 8.0+ (or MariaDB 10.5+).

• To upgrade your installation, follow these instructions.

• NOTE: The branches in the cerb/cerb-release repository have changed. The master branch has been removed, and new branches exist for each major version (e.g. v11.0, v11.1).

Added

• [Automations/LLM] In automations, added an llm.agent: command to simplify integrations with generative large language models (LLMs). The agent automatically manages history and tool use. Tools can be created using llm.tool automations with automatic documentation based on inputs:. The llm.agent: command will automatically invoke tools and return their output to the model. Integration is included for Anthropic, OpenAI, Groq, and Ollama.

• [Automations/LLM] In the llm.agent: automation command, non-automation tools: can be defined as type tool: with a description and parameters. When these tools are used by a model, the on_tool: event is triggered with custom logic and a __tool placeholder object with keys id, name, and parameters. A result must be returned using the tool.return: command. For example, this can use await:interaction: to delegate to any interaction, or to implement "human in the loop" manual approval workflows.

• [Automations/LLM] Added an llm.tool automation trigger. A collection of these tools can be provided to an llm.agent: using a large language model (LLM) that supports function calling (tool use). The inputs of an llm.tool automation are used to automatically generate the JSON Schema expected by common models. For instance, tools can provide access to Cerb data or integrate with third party APIs.

• [Interactions/Worker] In worker interactions, added a new llmTranscript form element. In two lines of code, this automatically handles rendering a chat history from an LLM session ID, with copy-to-clipboard, thumbs up/down ratings, etc. Previously, this had to be tediously implemented using a sheet element.

• [Interactions/Website] In website interactions, added a new llmTranscript form element. In two lines of code, this automatically handles rendering a chat history from an LLM session ID, with copy-to-clipboard, thumbs up/down ratings, etc. Previously, this had to be tediously implemented using a sheet element.

• [Automations/LLM] In automations, the llm.agent: command now returns a session_id to its output key. This can be used with the llmTranscript form element to render the chat history.

• [Automations/LLM] In automations, refactored the llm.agent: command so all tool use (including automations) executes the on_tool: branch. The same __tool placeholders are available as for custom tools. This allows real-time updates to be returned to the user during tool use; resetting the 30-second HTTP request time limit. Long-running deep research is now possible.

• [Devblocks/Platform/LLM] Added an LLM service to the Devblocks platform. This interfaces with Large Language Model APIs for generative text (e.g. Q&A, summarization, classification, report generation, email drafts).

• [Platform/LLM/Ollama] Added an ollama provider to the LLM service. This can connect to locally hosted open-source large language models through the Ollama API.

• [Platform/LLM/OpenAI] Added an openai provider to the LLM service. This can connect to hosted large language models through the OpenAI API (e.g. gpt-4o). A customizable api_endpoint_url also allows the use of other compatible services.

• [Platform/LLM/Groq] Added a groq provider to the LLM service. This connects to fast, hosted, open-source large language models through the Groq API.

• [Platform/LLM/Anthropic] Added an anthropic provider to the LLM service. This connects to hosted large language models through the Anthropic API (e.g. claude-3-5-sonnet-20241022).

• [Platform/LLM/HuggingFace] Added a huggingface provider to the LLM service. This uses the Serverless Inference API by default, but Inference Endpoints are also supported.

• [Platform/LLM/TogetherAI] Added a together provider to the LLM service. This uses the together.ai inference cloud service.

• [Automations/LLM] Added an llm.embedding automation trigger. This interfaces with large language model APIs to produce vector embeddings for blocks of text. For instance, embedding knowledgebase articles and FAQs for semantic search.

• [Automations/LLM] Added an llm.embed: command to automations. This runs llm.embedding automations to produce vector embeddings for blocks of text. For instance, embedding knowledgebase articles and FAQs for semantic search.

• [Automations/Inputs] In automations, each input can now set an optional description: key. This is used for documentation, autocompletion, and JSON Schema in LLM tool use.

• [Automations/Inputs] In automations, text: inputs can now set an optional allowed_values: key to create picklists. This is used for documentation, autocompletion, and JSON Schema in LLM tool use.

• [Devblocks/Platform/LLM] Added a 'DatabaseHistory' memory module to the LLM service for managing large language model chat history + context.

• [Automations/LLM] In automations, llm.agent: transcripts are now viewable in Setup->Developers->LLM Agent Transcripts. Transcripts can be marked read or deleted. When an agent uses a tool, its parameters and results are shown in the transcript. This makes debugging and compliance much easier. It also provides valuable feedback for improving agents based on past user interactions.

• [Automations/LLM] In LLM Transcripts, a 'Permalink' button provides a direct link for sharing transcripts.

• [Automations/LLM] When viewing LLM agent chat transcripts, metadata is displayed at the top (e.g. provider, user, client IP, automation, and automation node).

• [Automations/LLM] When viewing LLM agent chat transcripts, added a 'Copy' button to copy the plaintext (Markdown) content from a message to the browser clipboard.

• [Interactions/Website] In interaction.website automations, submit: form elements can now specify the width size: of each button: whole, half, third, and quarter.

• [Interactions/Website] In interaction.website automations, say: form elements can now specify several styles: options: text-center, text-large, text-left, text-right, or text-small.

• [Worklists/Search/Performance] Improved the performance of many complex search queries by running some subqueries independently and merging by IDs. This is enabled by default and can be disabled from the APP_OPT_SQL_SUBQUERY_TO_IDS configuration option.

• [Interactions/Website] In website interactions, the submit: form element has a new is_automatic@bool option. When true, the form is automatically submitted after it is rendered. This is particularly useful before a time-intensive operation like LLM text generation, which will instantly transition to a waiting spinner.

• [Interactions/Worker] In worker interactions, the submit: form element has a new is_automatic option. This is now preferred to await:duration: since it can render other form elements during the wait (e.g. say, sheet, LLM transcript).

Changed

• [Portals/Interactions] In 'Website Interaction' portals and on external websites, interactions now start in 'full' mode and use a blur overlay on the underlying website. The previous 'popup' default style was difficult to see on some website themes.

• [Worklists/Fieldsets] In worklists, improved the performance of the fieldset: filter.

• [Interactions/Website] In website interactions, the sheet: form element now applies styling to code blocks in markdown columns.

• [Setup/Plugins] In the setup page, moved 'Plugins' from its own menu into the 'Configure' menu. This menu had a single item, and we can use the space for something else.

• [Platform/Markdown] In Devblocks, changed the Markdown parsing library from Parsedown to League\CommonMark. Parsedown has some rendering issues and appears to have been abandoned since 2019.

• [Interactions/Worker] In worker interactions, continuing no longer hides the current step and only hides the submit button. The spinner is displayed in its place. This allows the auto-submit functionality to show progress messages.