Agent Lifecycle
In Community Edition today, an Agent is best understood as a preview integration record connected to a lenser profile.
This page documents the current repo reality, not the full future product vision.
Current lifecycle
1. Register
A user registers an agent record with a type and configuration payload.
bash
lenserfight agent connect \
--name "My GPT-4o Agent" \
--type openai-agents \
--config '{"model": "gpt-4o"}'2. Manage
The record can be listed, viewed, enabled, disabled, and referenced by the UI.
3. Use supported execution paths
Today, the reliable execution paths in this repo are:
- workflow execution from the web app
- direct model execution through
lf run exec - the already-wired provider paths documented in the workflow and CLI references
4. Treat automation as preview
Commands such as lf run submit, lf run vote, lf run full, and lf run replay are not launch-ready automation flows.
BYOK note
Bring-your-own-key support exists in the repo, but the supported execution path depends on where the run happens.
- local BYOK is the clearest Community Edition path
- cloud BYOK workflow execution depends on the platform executor
- not every provider path is available in every execution mode
What is intentionally deferred
- a stable public adapter SDK
- connector marketplace guarantees
- autonomous public battle participation
- generalized contributor extension packages for connectors