#We're building conversational AI agents using your Python SDK and have encountered an issue with wo
1 messages · Page 1 of 1 (latest)
AI-generated Answer for lacabracadabra
Short answer
- The docs state that workflows support both forward and backward edges; there is no published requirement that a workflow must be a DAG. The “Edges and flow control” section explicitly lists Backward Edges. (elevenlabs.io)
- The docs do not document any restriction against having edges in both directions between two nodes, nor do they mention a “duplicate edge” validation. I can’t find this limitation in the public docs. (elevenlabs.io)
Detailed answers
- Is this a known limitation / DAG constraint?
- Not documented. The Workflows guide describes Forward Edges and Backward Edges and does not say the graph must be acyclic. So, loops are supported conceptually, and no DAG enforcement is documented. (elevenlabs.io)
- Recommended pattern for conversation loops
- The documentation’s pattern for returning to earlier parts of the flow is to use Backward Edges with a condition (LLM, expression, or none). It doesn’t recommend inserting intermediate nodes as a special workaround. You can also control evaluation using edge ordering on a node. (elevenlabs.io)
- Can bidirectional edges be added after creation?
- Yes, workflows are part of the agent configuration and can be supplied/updated programmatically:
- On creation, include a workflow object (with edges and nodes) in the request. (elevenlabs.io)
- After creation, update the workflow via the Update Agent endpoint (workflow is included in request/response schemas). If versioning is enabled, make changes through Drafts (POST /v1/convai/agents/:agent_id/drafts) and then commit. (elevenlabs.io)
- Any undocumented workflow constraints to be aware of?
- The public docs do not list constraints about “duplicate edges,” bidirectional edges, or maximum numbers of edges/nodes. Documented, relevant details include:
- Edge types and conditions (LLM, expression, none). (elevenlabs.io)
- Node-level edge evaluation order via edge_order. (elevenlabs.io)
- Tool nodes have a special success/failure edge configuration. (elevenlabs.io)
What this means for your issue
- Your “duplicate edge” validation error is not covered in the published documentation. If you need an official confirmation or a fix, please open a ticket via the Help Center and include the payload causing the error. (elevenlabs.io)
References
- Workflows (edges, backward edges, tool-node edge behavior): (elevenlabs.io)
- Create Agent (include workflow on create): (elevenlabs.io)
- Update Agent (workflow in update schema): (elevenlabs.io)
- Drafts API (modify workflow when versioning is enabled): (elevenlabs.io)
- Agent Versioning (workflow is versioned; drafts/branches): (elevenlabs.io)
- Changelog
API has validation bug treating valid edges as duplicates