Persist Composed Agents as New Agents

[Biarritz1959]

When creating an agent and run it and then click on return to pattern, youreturns to the home screen and the agent definition is lost

[matoushavlena]

Hi @Biarritz1959, this is an intended behavior at this moment. The compose playground doesn't persist the composition configuration. Cc @anafucs @geneknit

[QuantGeekDev]

I want to see this feature too. I can add it. What are we thinking in terms of persistence? Keep it in the user's browser a la bolt.diy? Add postgres and docker compose file to beeai-server? Sqlite for a smaller footprint?
This would also involve UI changes - the current ux is quite streamlined, two choices - of which only one is selectable and the other is coming soon. We might want to add a new screen to display the user's saved workflows. Instead of "Start Composing", we might want to have "Next", and in the next screen there is a list of the saved workflows and a "New Workflow" button which will have the functionality that the "Start Composing" button currently has. We can start by saving with a "Save" button, and then implement the ability to save automatically on content change with a debounce

[QuantGeekDev]

I have a WIP implementation of this, will create a PR soon. Getting an error in the main branch related to the sequential workflow agent docker image crashing on start. Might be related to the acp-sdk, I see there's a branch in progress that might fix that. I added postgres, alembic, and sqlalchemy for db orm and migrations. The implementation is similar to my previous messages

Here's a sample from the DB in the new workflows table - the actual steps are in jsonb format:

"id"	"name"	"description"	"steps"	"created_at"	"updated_at"
"f9d08e6b-fa17-4fbb-9752-fced71057d6f"	"test"	"123321"	"[{""agent_name"": ""agent-docs-creator"", ""instruction"": ""test1""}, {""agent_name"": ""aider"", ""instruction"": ""test2""}]"	"2025-04-21 02:40:10.14456"	"2025-04-21 02:40:10.14456"`
``

[QuantGeekDev]

I get this error with the latest container for sequential-workflow:

docker run ghcr.io/i-am-bee/beeai-platform/official/sequential-workflow:latest
   Building sequential-workflow @ file:///app
      Built sequential-workflow @ file:///app
Uninstalled 1 package in 18ms
Installed 1 package in 0.72ms
Traceback (most recent call last):
  File "/app/.venv/bin/server", line 4, in <module>
    from sequential_workflow.server import server
  File "/app/src/sequential_workflow/server.py", line 6, in <module>
    add_sequential_workflow_agent(server)
  File "/app/src/sequential_workflow/agent.py", line 66, in add_sequential_workflow_agent
    @server.agent()
     ^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.11/site-packages/beeai_sdk/providers/agent.py", line 169, in decorator
    output = get_args(output)[0]
             ~~~~~~~~~~~~~~~~^^^
IndexError: tuple index out of range

[QuantGeekDev]

If anyone is running into the same, I've been able to work around this for now by using ghcr.io/i-am-bee/beeai/official/sequential-workflow:agents-v0.0.72 in agent-registry.yaml for the sequential-agent, and modifying the url resolution logic to point to the agent registry