Build together
A useful next step for every role
A dependable agent takes more than implementation. Use these guides to make product decisions, design conversations, evaluate outcomes, and prepare a release your team can operate.
Test whether your bank assistant uses the right account
Check that a bank assistant answers the customer’s request, even when two accounts return the same transactions in your test data.
Read the guide →Build a bank voice assistant with Deep Agents and Rasa Mantle
Connect Deep Agents research to a Rasa Mantle voice assistant. Learn how to validate evidence, handle corrections and bind approval to the right card charge.
Read the guide →How to set API keys after upgrading to Rasa 3.21
After a Rasa upgrade, write each model API key as an environment variable reference, and make sure the variable is set.
Read the guide →Choose a bank account with a local model and Rasa
Run a local model to suggest an account, and have Rasa ask the user to confirm it before showing transactions.
Read the guide →How to audit the documents an AI agent writes
List every way the model can change a document your agent writes, then test each one the way the agent calls it.
Read the guide →How to build an agent test set that exposes failures
Build an agent evaluation set with explicit task slices, traceable data, factual assertions, and failure cases that a headline success rate would hide.
Read the guide →Connect a LiveKit voice room to Rasa
Use LiveKit for the voice connection and speech services, and Rasa for dialogue, while checking messages after a participant reconnects.
Read the guide →How to choose a useful first task for an AI agent
Decide which workflow deserves an AI agent using a worked scope brief, a measurable outcome, and a boundary your product and engineering team can test.
Read the guide →How to stop an AI agent acting on an outdated confirmation
When an order or offer changes after the customer says yes, have the tool check that the terms read out still hold before acting.
Read the guide →How to stop curly apostrophes adding SMS segments
If your agent writes curly apostrophes, short SMS replies can take two segments instead of one; count them and replace the character before sending.
Read the guide →How to build test fixtures from your agent's own code
If your agent tests pass on a session you typed by hand, build that session from the key lists your agent's code reads and writes.
Read the guide →Human handoff from an AI agent: ask before sharing context
Design and test an agent-to-human handoff with explicit consent, a useful context packet, and recovery paths when the receiving team is unavailable.
Read the guide →How to stop customers repeating themselves after a handoff
Build the human agent’s first screen from the fields your voice agent collected, and test the transfer promise against what your own tools send.
Read the guide →How to set the model API key in Rasa endpoints.yml
Give every Rasa model group its API key through an environment variable placeholder, then start the server once against the real provider before you deploy.
Read the guide →Why TTS failover comes back after a voice agent restart
A restart wipes the voice router's memory of a failing vendor, and a total outage logs no failover, so alert on the failure type.
Read the guide →Build your first Rasa agent
Run a small Rasa agent locally, inspect the tool behind its answers, and make your first change with one complete downloadable starter.
Read the guide →Build a card-policy assistant with LangChain and Rasa
Use your existing document search to answer card-policy questions, with Rasa asking for details and explaining the sources.
Read the guide →Add a Rasa chat interface to a LangGraph workflow
Let users ask for a bank-statement summary through Rasa while LangGraph runs the calculation and saves unfinished work.
Read the guide →What a voice agent should say before verifying the caller
Caller ID tells your voice agent which number is calling, not who is holding the phone, so plan what it says before the caller confirms.
Read the guide →Help customers prepare a claim with CrewAI and Rasa
Use CrewAI to research the documents needed for an insurance claim, and Rasa to ask customers questions and explain the checklist.
Read the guide →Connect an Agno shopping adviser to Rasa
Help returning shoppers use their saved preferences through a Rasa assistant, while keeping each customer’s information separate.
Read the guide →Troubleshoot a slow connection with Strands and Rasa
Use your Strands network tools through Rasa to ask which area has a problem, explain the observations and keep callers’ histories separate.
Read the guide →Add Rasa without mixing up customer data or voice replies
Choose a working example for your existing agent or voice framework, and see which jobs Rasa can handle alongside it.
Read the guide →Test a delivery agent's handoff after a refused change
Inspect saved tool calls and unfinished skills to check what your delivery assistant does after a tool refuses a request.
Read the guide →AI agent release criteria: why 90% completion is not enough
Turn agent evaluation results into a release decision with an explicit denominator, a worked decision record, and owners for rollout and unresolved risks.
Read the guide →Why Rasa's memory discovery saved no facts on Claude
Why a Rasa agent on Claude answered a customer well but saved none of the facts they stated, and how to check yours.
Read the guide →How to set the embedding model for Rasa Mantle references
If your Rasa Mantle agent should embed its reference documents with a model you chose, name that model in the agent settings.
Read the guide →How to measure an AI agent handoff to a human
Test that each fact your human desk needs reaches the handoff on the agent’s own path, and measure handle time from real calls.
Read the guide →How to move a LangGraph or Strands agent to Rasa
Decide whether to move a LangGraph or Strands voice agent to Rasa, then port it step by step without losing its safety checks.
Read the guide →Migrate a Deep Agents database assistant to Rasa Mantle
Move a read-only database assistant to Rasa Mantle while keeping its Python query tools. Test the tool bindings, returned data and final answer before switching clients.
Read the guide →How to mutation test an AI agent guard in Python
Break your AI agent's guard on purpose, one small edit at a time, and check that your tests fail for every edit you make.
Read the guide →How to test a Rasa output hook on confirmed tool calls
A tool that asks the customer to confirm reaches your Rasa hook under another name, so test the hook with what it really receives.
Read the guide →Why a Rasa voice agent on Deepgram Flux can miss a short yes
When Deepgram Flux opens and closes a turn with no interim update, the Rasa handler commits nothing, so the agent never hears the reply.
Read the guide →How to redact spoken card numbers in a Rasa voice agent
If callers read card numbers to your Rasa voice agent, remove them in the speech-to-text engine before Rasa records the transcript.
Read the guide →Handle voice interruptions with Pipecat and Rasa
Connect Rasa to your Pipecat speech services, and discard an old answer when the caller interrupts and starts a new turn.
Read the guide →How to test an AI agent’s permissions
Review an agent’s permissions with a worked action register, evidence from the tool boundary, and named owners for authorisation, handoff, and unresolved risk.
Read the guide →AI agent rollout: rehearse the rollback before the pilot
Plan an agent rollout with dependency checks, outcome signals, a verified fallback, and a rollback drill that proves who can stop an unsafe release.
Read the guide →Is an MCP server a drop-in replacement for your tools?
Before you approve a move of agent tools to an MCP server, check where each tool now gets the customer id it acts on.
Read the guide →How to handle misheard names in a voice agent
When your voice agent gets numbers right but fails calls on names, measure names separately and list the known ones for speech to text.
Read the guide →What evidence to ask for before approving an agent action
A clean transcript does not prove an AI agent action is safe, so ask for tool tests that fail when the check is removed.
Read the guide →How to replace skill-level requires in Rasa
If a Rasa upgrade rejects the requires line in a skill, move the check into a precondition and delete any instruction that repeats it.
Read the guide →How to join Speechmatics segments into one Rasa turn
If your Rasa voice agent asks again for details the caller just gave, check whether your speech-to-text engine sends each phrase as its own turn.
Read the guide →Test tool confirmation in a Rasa agent that uses MCP
Run the same local tool request with confirmation enabled and disabled, then inspect whether the operation waited for approval.
Read the guide →How to make sure your agent sends the confirmation number
If your LLM agent completes an action but the customer never sees the reference, have the tool send it in the same turn.
Read the guide →How to measure voice agent latency to the answer
If your voice agent’s latency looks good but callers still wait, time each turn to the message that answers them, not to the first sound.
Read the guide →Stop an AI agent offering deals to opted-out customers
If your agent offers deals to customers who opted out, check consent inside the offer tool and stop tool results inviting the offer.
Read the guide →Ready to implement? Follow the Rasa tutorials for working examples and code.