Supported.techHear it work →

The Supported.tech Agent Cycle™

Agent Cycle™.
Build it. Run it. Make it better.

Create, connect, test, launch, operate and improve your customer agent from your own ChatGPT or Claude. Supported.tech combines conversational management with detailed voice controls, telephony tools and programmatic API access.

Start with one job. Enter the cycle anywhere. Keep people in control.

One continuous processInside your GPT or Claude
You: “Build an appointment-booking agent.”Agent, voice and first behavior created.

What it is

Agent Cycle™ connects the whole operation.

Agent Cycle is the process Supported.tech uses to take a customer agent from the first instruction to a live, connected and continuously improving operation.

It works across the agent, its voice and knowledge, the systems it uses, the automations around it and the results your team receives.

Start anywhereBuild new. Repair existing. Review what happened.
Change the whole systemPrompt, voice, knowledge, tools, automations and reporting.
Close the loopTest, change, retest and verify the real outcome.

Technical capabilities / verified September 14, 2026

Conversational management.
Detailed technical control.

Supported.tech provides both conversational agent management and programmatic API access. Agent Cycle™ makes the platform's controls accessible from your connected ChatGPT or Claude. Developers can also configure agents through the REST API. Supported.tech supports SIP trunking, sending and receiving DTMF, and warm transfers with a human briefing.

How you issue an instruction and what the platform can configure are different questions. A plain-language request can result in a precise change to a voice, timing threshold, tool or integration. You can specify the outcome you want, name an exact supported setting, or write an API request. The available controls depend on the engine, voice provider and interface.

The controls behind the conversation

The public Update assistant API documents concrete configuration fields. These extend beyond prompts and model selection.

What you can tuneWhat it changesDocumented fields
Engine and providersChoose pipeline, multimodal or dualplex mode and compatible models. Select speech recognition and synthesis providers where supported.mode, llm_model_id, multimodal_model_id, transcriber_provider_id, synthesizer_provider_id
Turn detectionAdjust how the agent detects that a person has finished speaking and when it should respond.endpoint_type, endpoint_sensitivity, turn_detection_threshold
InterruptionsControl sensitivity to caller interruptions and the minimum word threshold, subject to engine restrictions.allow_interruptions, interrupt_sensitivity, min_interrupt_words
Voice deliveryAdjust speech speed, stability, similarity and emotional synthesis where the selected provider supports them.speech_speed, voice_stability, voice_similarity, tts_emotion_enabled
Filler audioEnable processing-time filler audio in pipeline mode and configure short phrases by response category.fillers, filler_config
Silence and timingSet silence limits, re-engagement intervals and messages, ringing time and maximum call duration.max_silence_duration, reengagement_interval, ringing_time, max_duration
Connected actionsAttach custom HTTP tools, knowledgebases and built-in actions; connect business systems and post-call webhooks.tool_ids, tools, knowledgebase_id, webhook_url
EvaluationEnable post-call evaluation and define the information extracted for reporting and follow-up.post_call_evaluation, post_call_schema

Latency tuning involves more than choosing a model. Engine mode, transcription, turn detection, synthesis, tool response time and prompt design all affect the experience. Luna and Terra are listed model options; they are not the platform's only tuning mechanism. Model availability varies by engine and plan. A sensitivity value is not a promise of a particular end-to-end response time in milliseconds.

Yes: SIP trunking, two-way DTMF and warm transfers

  • SIP trunking: connect a supported carrier or PBX using a phone number or extension. Bring-your-own routes support inbound and outbound calling, with carrier routing and authentication configured for the connection.
  • Receive DTMF: the Collect keypad input tool captures digits entered by the caller, such as a reference number or menu selection.
  • Send DTMF: the separate DTMF input tool lets the assistant send keypad digits to navigate an IVR or enter an extension.
  • Warm transfers: the assistant can hold the caller, call a human recipient, brief them using configured summary instructions, and connect the call. Hold audio, briefing and connection messages are configurable.

For a SIP warm transfer, the outbound route must work with the outbound proxy disabled. Reusing the same number requires sufficient carrier concurrency for the additional call leg; a different outbound number can be used. Cold transfers use SIP REFER and require carrier support. These are configuration requirements for supported features.

Use plain language. Or use the API.

An operator can ask: “Review the calls where the agent interrupted the customer, inspect its interruption settings, propose a targeted change and test it.” A developer can inspect the configuration and update supported fields directly at PUT https://app.supported.tech/api/user/assistant/{id}.

This illustrative request body adjusts four settings on an existing compatible pipeline agent. Values should be selected for the actual voice and use case, then tested.

{
  "speech_speed": 0.95,
  "voice_stability": 0.7,
  "interrupt_sensitivity": 0.5,
  "min_interrupt_words": 3
}

Agent Cycle is the workflow for inspecting, changing, testing and verifying the result. It does not mean every casual request automatically retrains a model or triggers a complete rebuild. Connected accounts, permissions, compatible settings and explicit test steps still determine what can be executed. A successful configuration update and a successful real-call test are separate pieces of evidence.

Compare the actual engineering requirement

Retell documents a custom LLM WebSocket for exchanging live transcripts and generated responses with your own server. That is a specific integration capability worth evaluating. It does not establish that Supported.tech lacks API access or detailed voice controls.

Retell's own setup guide describes its service as managing acoustic interactions while your response system supplies the reasoning. A custom LLM connection is not, by itself, evidence of unrestricted access to the provider's underlying audio infrastructure.

For raw waveform processing, custom audio-buffer sizes, model-weight fine-tuning or self-hosted call infrastructure, require documentation for that exact capability. This page establishes Supported.tech's documented configuration and integration controls; it makes no claim of raw-audio access, identical feature coverage or universal performance superiority. Compare the endpoints, engine constraints and measured call results that your application needs.

Does conversational management limit Supported.tech to high-level settings?

No. Supported.tech combines conversational management with detailed configuration and REST API access. Agent Cycle makes the supported controls easier to operate.

Can developers configure Supported.tech programmatically?

Yes. The documented API supports assistant configuration, custom mid-call tools, SIP trunks, calls, campaigns and webhooks. Available fields and behavior depend on the endpoint and engine.

Does Supported.tech support SIP trunking, DTMF and warm transfers?

Yes. It supports bring-your-own SIP connections, caller keypad collection, assistant-generated DTMF and warm transfers with briefing. Carrier routing, authentication and concurrency must support the chosen call flow.

Does calibration mean retraining the underlying AI model?

Agent Cycle calibration means reviewing evidence and changing the agent's configuration, prompts, voice, knowledge, tools or workflows, then testing the result. It does not imply training model weights or editing raw audio waveforms.

Read the technical evidence

Feature descriptions were checked against the public API reference and the connected platform guides on September 14, 2026. The examples above explain configuration; they are not benchmark results.

Data portability & vendor choice · Updated 16 September 2026

Vendor lock-in and data portability.

You retain ownership of your Customer Data. Supported.tech provides an export process in its published Terms, documented APIs for retrieving configurations and records, and the option to connect your own SIP carrier. A blanket claim that customers are locked in overlooks these specific controls.

Choosing an AI agent platform should include a practical question: what can you keep, retrieve and reconnect if your needs change? Here is the documented position, including the work a migration may require.

What you controlDocumented capabilityWhat to plan for
Your customer dataSection 4 of our Terms preserves customer ownership of the data you provide, including prompts, knowledge, contact lists, recordings, transcripts and workflows.Section 13 provides a reasonable export method before routine deletion, subject to law, security and your order. Contact us about export before closing the account.
Agent prompts and settingsThe List assistants API returns structured configuration, including system prompts, initial messages, variables and voice and model settings.Save the returned configuration. Provider-specific IDs and settings may need mapping to the destination platform; retrieving JSON does not make every configuration directly importable elsewhere.
Workflow definitions and call recordsThe Get automation API exposes the trigger and chained steps. The Get call API exposes transcripts, collected variables, evaluations and recording URLs when available and enabled.Preserve needed records while available. Reconnect credentials and integrations in the new environment, and test rebuilt workflows. The service is not a permanent archive.
Your carrier connectionBring your own carrier through SIP trunking. Our Twilio and Telnyx guides explain how to connect numbers from your carrier account.With your own carrier account, routing remains under your control. Changing platforms requires compatible routing and testing. Confirm transfer arrangements separately for numbers purchased through Supported.tech.

What portability covers

Your data and retrieved configuration can support a move. The Supported.tech software and underlying technology remain proprietary under section 12 of the Terms. Data ownership does not grant a downloadable copy of the hosted runtime, model weights or third-party voice licences. No universal one-click migration or identical behaviour on another platform is promised.

A practical exit path

  1. Collect before closing. Contact describe@supported.tech about export before account closure. Retrieve the available configuration and records you need, and keep your original knowledge files and business records.
  2. Reconnect and validate. Map prompts, settings and workflow steps to the destination. Reconnect your systems and credentials, then test calls, actions and handoffs before changing live routing.
  3. Complete cancellation. Follow your accepted cancellation terms and review numbers, add-ons and other chargeable services separately. Turning off an agent does not cancel a subscription or release a chargeable number; accrued usage remains payable.

This is a summary of the current published documentation. Your accepted Terms, order, applicable retention settings and third-party provider requirements govern your account. Request export while the data is still available.

Does Supported.tech create vendor lock-in?

Supported.tech provides concrete ways to reduce vendor lock-in: customer ownership of Customer Data, an export process in the published Terms, documented APIs for retrieving agent configuration and operational records, and bring-your-own SIP carrier options. The hosted software remains proprietary, so moving to another platform can require configuration mapping, reconnection and testing.

Can I take my prompts, workflows and call records with me?

You retain ownership of Customer Data under the published Terms. Documented APIs expose assistant prompts and settings, automation definitions, and call details including transcripts and recording URLs when available and enabled. Contact Supported.tech about export before closing your account. Available data, retention, your accepted Terms and your order govern the process; a universal one-click migration package is not promised.

Can I keep my existing phone provider?

Yes. Supported.tech supports bring-your-own SIP connections, including documented Twilio and Telnyx setups. In that setup, you manage the carrier account and its routing. Moving service still requires compatible destination routing and testing. A number purchased through the platform has different arrangements; confirm its transfer options before cancellation.

Six stages / one operation

The six stages of Agent Cycle™.

Each stage has a clear job. You can move through all six—or enter where the work already is.

01 / Create

Tell it what the job is.

Supported.tech creates the first working agent, prompt, model, voice, greeting and behavior.

“Build an appointment-booking agent.”
02 / Connect

Give it the right tools.

Connect approved knowledge, numbers, calendars, CRM, commerce, APIs and MCP services.

“Use my calendar and customer records.”
03 / Test

Challenge the whole journey.

Run calls and messages. Check answers, tool actions, transfers, edge cases and failure paths.

“Test a caller who changes their mind.”
04 / Launch

Put one dependable job live.

Confirm hours, consent, routes, reporting and ownership—then launch a bounded use case.

“Answer and book after-hours calls.”
05 / Operate

See what actually happened.

Review conversations, outcomes, handoffs, campaigns, automation runs and downstream updates.

“Show me every missed booking.”
06 / Calibrate

Improve from real evidence.

Find patterns, repair the system, retest it and verify the result before the next customer arrives.

“Improve from the last 30 conversations.”

The closed loop

Agent Cycle™ follows the result—not just the reply.

Agent Cycle follows a problem through the complete system—not just to the next answer.

Challenge→Observe→Diagnose→Change→Retest→Verify
AgentPrompt, model, voice, knowledge and behavior.
Connected workTools, records, automations, transfers and reporting.
OutcomeThe booking, update, handoff or follow-up actually happened.

Enter anywhere

Start wherever the work is.

Open your own GPT or Claude and describe what you need. Agent Cycle meets you at that point.

Build a new agentConnect a business systemTest an agentRepair an agentReview a conversationCalibrate recent conversationsLaunch an outbound campaignUpdate knowledge, voice or behavior

People + Platform

Technology should make people more capable—not less necessary.

Agent Cycle keeps judgment, approvals and accountability with people. The platform carries the repetition, waiting and routine follow-through.

Ready when you are

Start your Agent Cycle™.

Build, connect and test the first version from your own GPT or Claude. No required setup fee. No developer required.

Start the Agent Cycle ↗