canonical: https://jentic.com/apis/messagebird.com/messagebird-intent-api # Messagebird Intent Jentic publishes the only available OpenAPI specification for Intent, keeping it validated and agent-ready. The MessageBird Intent API classifies short text utterances into customer-defined intents using trained classification models that live in datasets. Agents send a phrase to a chosen dataset and receive predicted intent labels with confidence scores, which is useful for routing inbound SMS, WhatsApp messages, or chatbot turns to the right workflow. Supported languages are exposed via a dedicated endpoint, and dataset metadata reveals training status before predictions are attempted. ## For AI agents Classify customer messages into trained intent labels by calling a MessageBird Intent dataset, with confidence scores returned per prediction. ## Scope Does not handle message delivery, language detection, or entity extraction - use for intent classification of short text only. ## Capabilities - Classify an inbound SMS phrase into a trained intent label and read its confidence score - List all intent classification datasets available on the account before choosing one for prediction - Inspect a single dataset to confirm its training status before sending production traffic - Look up the languages supported by the Intent service to gate predictions on unsupported locales - Route customer support replies into automated workflows based on the predicted intent label ## Use cases ### Inbound Support Triage Classify inbound SMS or chat messages into a fixed set of support intents (billing, cancellation, technical issue) so each conversation is routed to the correct queue or automated reply. The Intent API runs the prediction against a dataset that was previously trained on labelled examples and returns the top intent label with a confidence score, letting the workflow fall back to a human when confidence is low. Triage decisions can be made in milliseconds before the message ever reaches an agent. Example prompt: POST a customer message to /api/v1/datasets/{datasetId}/predict on the support classifier dataset and forward it to the billing queue when the predicted intent is 'billing_question' with confidence above 0.8 ### Chatbot Disambiguation Use predicted intents to drive branching logic inside an automated chat flow. When a user replies in free text, the bot calls the Intent API with the user's phrase and uses the returned label to pick the next step, instead of relying on brittle keyword matching. Datasets can be retrained as new intents emerge without changing the bot code, and the supported-languages endpoint lets the bot decline gracefully on unsupported locales. Example prompt: Call POST /api/v1/datasets/{datasetId}/predict with the user's chat reply and branch the flow based on the predicted intent label ### Dataset Readiness Monitoring Before promoting a new classifier into production traffic, agents can list all datasets and inspect each one individually to confirm training has finished. The dataset detail endpoint exposes status information so a deployment workflow can wait for training to complete before flipping traffic. This avoids the failure mode of sending live messages to a half-trained model. Example prompt: GET /api/v1/datasets and then GET /api/v1/datasets/{datasetId} to verify training has completed before routing live traffic to the classifier ### AI Agent Integration via Jentic An AI agent uses Jentic to discover the Intent API by intent search ('classify a customer message') without needing to read MessageBird's documentation. Jentic returns the predict operation with its input schema, so the agent can submit a phrase and receive a structured intent prediction in a single load-and-execute step. Credentials are isolated in your Jentic One instance and never enter the agent's prompt. Example prompt: Use Jentic to search 'classify a customer message into intents', load the Intent predict operation schema, and execute it against the production support dataset ## Key endpoints | Method | Path | Description | | --- | --- | --- | | GET | /api/v1/datasets | List all intent datasets | | GET | /api/v1/datasets/{datasetId} | Get a single dataset including training status | | POST | /api/v1/datasets/{datasetId}/predict | Predict the intent of a phrase against a dataset | | GET | /api/v1/supported-languages | List supported classification languages | ## Key resources - **Datasets** — Trained intent classification models scoped to an account, listed and inspected before use - **Predictions** — Intent predictions returned for a phrase against a chosen dataset, including confidence scores - **Supported Languages** — List of languages the Intent service can classify phrases in ## Why Jentic - **Setup:** Wiring the MessageBird Intent API by hand means attaching your AccessKey to each request against the intent.messagebird host, routing dataset lookups and the predict call, and handling retries yourself. Through Jentic you install once, import the Intent API from the API Directory, store the AccessKey once, and your agent calls it. - **Permission scoping:** The Intent API puts the dataset id in the URL path (/api/v1/datasets/{datasetId}/predict), so a rule can pin your agent to one dataset. You choose the operations it may call, so a rule can grant dataset reads and intent prediction while leaving out anything else. - **Credential handling:** Your MessageBird AccessKey is stored once, encrypted, by your own Jentic One instance and attached at execution time. It never enters the agent's prompt, logs, or context. - **Discovery method:** Agents search Jentic by intent such as 'classify a customer message', and Jentic returns the Intent predict operation with its input schema so the agent calls the right endpoint without browsing the reference docs. ## Related APIs - **MessageBird Language Detection** — Detect the language of a phrase before passing it to the Intent classifier - **MessageBird Named Entity Recognition** — Extract entities like names, dates, and locations from the same message after intent classification - **OpenAI API** — General-purpose LLM that can classify intents via prompting instead of a trained MessageBird dataset - **Cohere** — Hosted classifier and embedding API for building intent classifiers without managing infrastructure ## FAQ ### Why is there no official OpenAPI spec for Intent? MessageBird does not publish an OpenAPI specification for the Intent API. Jentic generates and maintains this spec so that AI agents and developers can call Intent via structured tooling. It is validated against the live API and kept up to date. Get started with Jentic One, the self-hosted execution layer. ### What authentication does the MessageBird Intent API use? The Intent API uses an API key sent in the `Authorization` header, prefixed with the literal word `AccessKey`. Through Jentic, the access key is stored in your Jentic One instance and injected at request time, so the raw key never enters the agent's context. ### Can I list available intent datasets before running predictions? Yes. Call GET /api/v1/datasets to enumerate every dataset on the account, then GET /api/v1/datasets/{datasetId} for full details on a specific one. This is how you confirm a model has finished training before sending live traffic to it. ### What languages does the Intent API support? Call GET /api/v1/supported-languages to retrieve the full list at runtime rather than hardcoding it. The set evolves as MessageBird adds language coverage to the underlying classifier. ### How do I classify a message through Jentic? Search Jentic for `classify a customer message into intents`, load the POST /api/v1/datasets/{datasetId}/predict operation, and execute it with your phrase and target dataset. Jentic returns the predicted intent label plus confidence in a structured response. ### Are there rate limits on the Intent API? The OpenAPI spec does not document specific quantitative rate limits, so check your MessageBird account dashboard or contact MessageBird support for the limits attached to your plan before bulk-classifying traffic. ### Can I limit what my agent is allowed to do with the MessageBird Intent API? Yes. Because you run Jentic One yourself, your own rules decide which Intent operations and credentials your agent may use. You can allow only dataset reads (GET /api/v1/datasets and GET /api/v1/datasets/{datasetId}) plus the predict call (POST /api/v1/datasets/{datasetId}/predict) and leave out anything else. Since the dataset id sits in the URL path, a rule can also pin the agent to a single classifier dataset so it cannot run predictions against others.