# Alternatives to Vapi > Source: https://theagentsindex.com/vapi/alternatives (current datasheets, evidence-ranked) Candidates share the baseline category or are previously researched peers. Requirements, preserved strengths, explicit trade-offs and research coverage decide the order; paid tier does not. ## Shortlist ### 1. ElevenLabs Agents Hosted platform for building voice and chat agents, with telephony, SDKs and a choice of LLM. **Research coverage:** 35/35 criteria **Where it improves on Vapi:** - [Agent permissions: yes — Tool use is gated per MCP server and per tool: always ask, fine-grained (auto-approved, requires approval, disabled), or no approval at all.](https://elevenlabs.io/docs/eleven-agents/customization/tools/mcp) - [In your pipeline: yes — Agents deploy from CI with an API key in the environment; the unit of change is the agent configuration, not a pull request in your code.](https://elevenlabs.io/docs/eleven-agents/operate/cli) - [Seat model: yes — Two seat types with a hard ceiling: Full Seats are capped at 11 per self-serve workspace, every plan includes 20 Basic Seats, and Enterprise seat counts sit in the contract.](https://elevenlabs.io/docs/overview/administration/workspaces/members) **Trade-offs against Vapi:** - [In your editor: no — The vendor enumerates its surfaces as web widgets, mobile apps, phone systems and custom integrations; no editor or IDE extension appears in that list.](https://elevenlabs.io/docs/eleven-agents/integrate/overview) - [Admin controls: partial — Guardrails constrain what an agent may say and per-tool approval constrains what it may do, but Guardrails is in Alpha and enabling MCP servers is not restricted to admins.](https://elevenlabs.io/docs/eleven-agents/best-practices/guardrails) - [Getting out: partial — Agents, tools and tests pull down as files via the CLI and conversations are readable over the API, but unused paid credits expire on cancel or downgrade and rollover caps at two months.](https://elevenlabs.io/pricing/agents) [View ElevenLabs Agents](https://theagentsindex.com/elevenlabs-conversational-ai) ### 2. PolyAI Enterprise platform for building voice and chat agents that resolve customer calls, priced per minute. **Research coverage:** 35/35 criteria **Where it improves on Vapi:** - [In your editor: yes — The ADK edits agent resources as local files in any editor, and Builder MCP plugs the same workflow into Cursor, Claude Code, Claude Desktop and Codex.](https://poly.ai/developers) - [In your pipeline: yes — poly push is CI-compatible and the Agents API lets a pipeline gate promotion on a passing simulation-test set.](https://docs.poly.ai/testing/ci-automation) - [Data residency: yes — Enterprise workspaces are pinned to a US, UK or EU region with matching API and MCP hosts; self-serve accounts live in the shared Studio region.](https://docs.poly.ai/mcp/quickstart) **Trade-offs against Vapi:** - [Multi-agent: no — The only 'team of agents' feature is assigning several TTS voices to one agent; no coordination between multiple agents on a single task is documented, and the build assistant works one branch at a time.](https://docs.poly.ai/voice-channel/multi-voice) - [Open models: partial — The model catalogue enumerates only Raven, OpenAI and Amazon Bedrock (Claude, Nova) with no open-weight option, but you can attach your own LLM endpoint.](https://docs.poly.ai/behavior/models/byom) - [Self-hosted: no — Runs only as PolyAI-hosted SaaS on AWS with region-specific workspaces; no on-premise or customer-cloud deployment appears anywhere in the vendor's docs or security pages.](https://poly.ai/security) [View PolyAI](https://theagentsindex.com/polyai) ### 3. Cartesia Real-time speech models and a managed runtime for production phone and web voice agents. **Research coverage:** 35/35 criteria **Where it improves on Vapi:** - [Seat model: partial — Seats are not priced, but inviting anyone requires Startup ($49/mo) or above; a Free organization is capped at one member. The pricing page's 'unlimited workspace seats on every plan' contradicts this.](https://docs.cartesia.ai/enterprise/set-up-an-organization) - [Certifications: yes — GDPR, SOC 2 Type II, HIPAA and PCI-DSS service provider, with a public Trust Center; DPAs and BAAs come with Enterprise.](https://docs.cartesia.ai/enterprise/zero-data-retention) **Trade-offs against Vapi:** - [Done-for-you service: no — Self-serve software: you build and deploy the agent yourself; sales contact exists for Enterprise terms, not for building it for you.](https://www.cartesia.ai/agents) - [Multi-agent: no — System tools enumerate three slots: end call, DTMF, and transfer to an E.164 phone number. An agent therefore hands off to a number, not to another Cartesia agent.](https://docs.cartesia.ai/agents/system-tools) - [Open models: partial — The LLM layer routes through LiteLLM to 100+ providers, but the vendor names only Anthropic, OpenAI and Google, and its own speech models are API-only.](https://docs.cartesia.ai/line/sdk/agents) [View Cartesia](https://theagentsindex.com/cartesia) ### 4. LiveKit Agents Open-source Python and Node.js framework for voice AI agents, with LiveKit Cloud hosting, telephony and observability. **Research coverage:** 35/35 criteria **Where it improves on Vapi:** - [Self-hosted: yes — Agents are containers you can run on your own Kubernetes; the LiveKit media server and SIP service are separately self-hostable. Self-hosted agents still meter WebRTC participant minutes if they connect to LiveKit Cloud.](https://docs.livekit.io/deploy/custom/deployments.md) - [Open source: yes — Apache 2.0 for the Agents SDK and for the media server, SIP, Egress, Ingress and client SDKs. LiveKit Cloud is the commercial hosted layer on top.](https://livekit.com/voice-agents) - [Opt out of training: yes — Nothing to opt out of: zero retention is the default on every plan, LiveKit logs no prompts, audio or completions, and voice-clone samples are opted out of provider training on your behalf.](https://livekit.com/security/overview) **Trade-offs against Vapi:** - [Done-for-you service: no — You run the software: an open-source SDK plus a self-serve cloud. Enterprise adds a designated solutions engineer, which is support, not a build-for-you engagement.](https://livekit.com/products/agent-platform) - [MCP server: partial — The agent is the MCP client: it consumes tools from remote HTTP or local stdio MCP servers, Python only. LiveKit does not expose a running agent as an MCP server for another agent to query.](https://docs.livekit.io/agents/logic/tools/mcp.md) - [Bring your own key: partial — Provider plugins run on your own account, key and billing, and self-hosted agents run in your own cloud. Bringing your own key to traffic routed through LiveKit Inference is not supported.](https://docs.livekit.io/agents/models/llm/) [View LiveKit Agents](https://theagentsindex.com/livekit-agents) ### 5. Retell AI Pay-as-you-go platform for building, testing and running AI phone, chat and SMS agents. **Research coverage:** 34/35 criteria **Where it improves on Vapi:** - [Certifications: yes — HIPAA, GDPR and SOC 2 Type 1 and Type 2 are claimed, with BAA and DPA self-signing at no extra fee and reports available through a trust centre.](https://docs.retellai.com/general/compliance) **Trade-offs against Vapi:** - [Open models: partial — Retell's own model menu is limited to hosted GPT, Claude and Gemini, but the custom-LLM path lets you put a self-hosted open-weights model behind the agent over a WebSocket you operate.](https://docs.retellai.com/integrate-llm/overview) - [On the command line: no — The docs enumerate every way to drive Retell: dashboard, REST API, Node/Python SDKs and MCP server. No command-line tool is among them.](https://docs.retellai.com/general/introduction) - [Self-hosted: no — The platform runs on Retell's own infrastructure: pay-as-you-go uses shared servers and the most isolation offered is a dedicated Retell server sold as an Enterprise add-on, with no customer-infrastructure deployment documented.](https://www.retellai.com/pricing) [View Retell AI](https://theagentsindex.com/retell-ai) ### 6. Telnyx Voice AI Agents Carrier-owned platform for phone and web voice agents, with telephony, STT, TTS, LLM and orchestration in one stack. **Research coverage:** 35/35 criteria **Where it improves on Vapi:** - [Audit log: yes — An audit events API returns who changed what and when, including whether the change came from Telnyx, the owner, an organization member or an account manager, though the vendor calls it best-effort.](https://developers.telnyx.com/api-reference/audit-logs/list-audit-logs) - [Single sign-on: yes — SAML single sign-on is configurable per organization with IdP entity ID, SSO target URL and certificate fingerprint; only the organization owner can reach the SSO settings page.](https://developers.telnyx.com/api-reference/authentication-providers/creates-an-authentication-provider) - [Certifications: yes — SOC 2 Type II is claimed on the product page, and GDPR, PCI and HIPAA readiness is presented as standard on every plan including pay-as-you-go rather than a paid enterprise add-on.](https://telnyx.com/products/voice-ai-agents) **Trade-offs against Vapi:** - [Done-for-you service: no — You build and run the agent yourself in the portal or over the API; Telnyx sells platform capacity, and human help appears only as a dedicated account manager on the $500-minimum Committed plan, not as a build service.](https://telnyx.com/products/voice-ai-agents) - [GPT: partial — GPT models run as an external provider you supply credentials for: you create an OpenAI API key, store it as an integration secret and select the model, and OpenAI free-plan keys are rejected outright.](https://developers.telnyx.com/docs/inference/data-residency) - [Self-hosted: no — The runtime, telephony and portal stay on Telnyx; only the LLM and TTS steps can be pointed at your own endpoint, and Enterprise 'dedicated infrastructure' is still Telnyx-operated.](https://developers.telnyx.com/docs/inference/ai-assistants/no-code-voice-assistant) [View Telnyx Voice AI Agents](https://theagentsindex.com/telnyx-voice-ai) ## Method Unknown data is never treated as a benefit or a trade-off. Featured placement is labelled but cannot alter recommendation order.