Clarification on Multi-Prompt Agent Legacy Status

Support Help
1

Subject: Clarification on Multi-Prompt Agent Legacy Status

Hi Retell Support,

We noticed that Multi-Prompt Agent is marked as Legacy in Retell. We are building production voice agents for a medspa use case, with flows such as appointment booking, rescheduling, cancellation, account updates, service questions, Booker API tool calls, caller verification, and human escalation.

Before we finalize our architecture, could you please clarify the following?

  1. Why has Multi-Prompt Agent been moved to Legacy?

    • Is this due to known reliability issues, latency issues, state handling problems, tool-calling issues, transition failures, or context loss?
    • Or is it simply because Conversation Flow is now the newer recommended builder?

  2. Is Multi-Prompt Agent officially deprecated?

    • If yes, is there a confirmed support timeline, cutoff date, or migration deadline?

  3. Are existing Multi-Prompt Agents safe to use in production for the next 6 to 12 months?

    • Will they continue to receive bug fixes, runtime improvements, model updates, and support?

  4. For a complex production voice agent with multiple workflows, should we build using Conversation Flow instead of Multi-Prompt Agent?

  5. Is there full feature parity between Multi-Prompt Agent and Conversation Flow for:

    • State or node-specific tools
    • Variable and context preservation
    • Function calling
    • Webhooks
    • Post-call analysis
    • Versioning
    • Simulation testing
    • API-based creation and updates
    • Phone number assignment
    • Agent transfer or handoff between specialist agents

  6. Is there an official migration guide or recommended process to move existing Multi-Prompt Agents to Conversation Flow?

  7. Are there any known limitations or failure patterns in legacy Multi-Prompt Agents that we should avoid, especially around:

    • Stuck states
    • Incorrect transitions
    • Tool-calling reliability
    • Latency
    • Context loss
    • Debugging difficulty

  8. For our use case, what architecture would Retell recommend?

    • One large Conversation Flow agent
    • Single Prompt plus Agent Transfer to specialist agents
    • Multiple specialist agents connected through Agent Transfer
    • Some other recommended pattern

Could you also share any official documentation, release notes, or roadmap notes explaining why Multi-Prompt Agent is marked as Legacy and what Retell’s long-term recommendation is?

Thanks,
Sabari

2

Hi @sabari.selvaraj Thanks for flagging — happy to clarify before you commit to an architecture.
Why “Legacy”? It’s a UI steering label we added when we introduced the new agent-type picker. It signals that Conversation Flow is the recommended builder going forward for new complex agents — not that Multi-Prompt has reliability, latency, tool-calling, or state-handling issues. There is no quality regression behind the label.
Is it deprecated? No. There is no published deprecation notice, no sunset date, and no migration deadline. The Multi-Prompt docs and product surface remain live, and the runtime is actively maintained — in fact, parts of it are shared infrastructure that Conversation Flow itself depends on internally, so it isn’t going anywhere in the foreseeable future.
Safe for production over the next 6–12 months? Yes. Existing Multi-Prompt agents continue to receive bug fixes, model updates, and runtime improvements alongside the rest of the platform. We’ll give meaningful advance notice through normal deprecation channels if that ever changes.
Multi-Prompt vs Conversation Flow for your medspa use case (booking, rescheduling, cancellation, account updates, service Q&A, Booker API tools, caller verification, human escalation):
For a workflow with that many distinct branches, deterministic transitions, and tool calls per state, Conversation Flow is the better fit. You get:

  • Visual node/edge editing of state transitions (easier to audit and iterate on with non-engineers)
  • Explicit per-node tool, prompt, and transition config — less prompt-engineering to keep states from bleeding into each other
  • Logic Split / Global Nodes for things like verification gating and escalation that are awkward to express in a single multi-prompt graph

Multi-Prompt remains a valid choice if you prefer pure prompt-driven state control, but for the scope you described Conversation Flow will scale more cleanly.

Thank You