At a glance
- Category
- Agent tools and integrations
- Tags
- No credit card
For developers
- Works with
- Vapi, Retell, Bland, LiveKit, Pipecat, ElevenLabs and 2 more
Zelto in our research
All researchStudy
AI Tool Pricing Index, November 2026
Pricing
Contact sales
82 of the 361 tools we checked ask you to contact sales for a price.
Edition price, checked 20 October 2026.
Pricing pages of the AI tools, November 2026 editionas of n = 361 tools checked
Zelto grades production calls instead of sampling them
Zelto sits between a voice agent platform and the team that maintains the agent. Conversations arrive from a connected provider, each becomes a turn-by-turn transcript with its recording re-hosted on Zelto storage so the audio outlives the provider's own retention, and analysis runs over every one of them. Recurring problems are promoted into findings, durable org-level issues with the exact transcript moments attached, and a change records the fix.
The unit of ongoing work is a monitor: a continuous check built from an ordered pipeline of stages. Turn-count, character-count and regex filters run first and cost nothing, then a tool check, a TypeScript code stage, an AI judgement or a flow check produces a typed value. One stage is designated the headline, and its result drives the affected rate you trend and alert on. Tool checks carry the sharpest rule on the page: confirmation language in the transcript is recorded as a claim, never as evidence that the tool ran, and a conversation with no tool metadata is marked insufficient evidence rather than a pass.
The marketing site is voice-first. The documentation is wider: chat, SMS, WhatsApp and web-chat conversations land on the same ingest endpoint with no audio.
Connecting a provider takes a key, not an SDK
Vapi, Retell and ElevenLabs connect by pasting an API key, after which Zelto pulls conversations on a sync, with an optional provider webhook for lower latency. LiveKit forwards each finished session from your worker, and everything else, Bland and Pipecat included, POSTs to /webhooks/calls. Cartesia, Telnyx, Speechify, Twilio for carrier-level traces and Kapso for WhatsApp each have their own page.
None of it changes how the agents run, which is the point of the design: there is no SDK in the call path. The vendor says most teams are sending traffic in under 15 minutes.
Outputs leave the same way they came in. Positive monitor results go out as signed monitor.matched webhooks with HMAC-SHA256 headers, at-least-once delivery, six attempts and a documented Node verifier. Findings can be filed as Linear issues, and alerts route to Slack channels or email on a daily sweep.
The MCP server is the most complete surface
Zelto runs an MCP server at api.zelto.ai/mcp that accepts either an org API key or a remote OAuth 2.1 sign-in with Dynamic Client Registration and PKCE. OAuth access tokens live one hour, refresh only while the Zelto session is valid, and are scoped to the single organization chosen at consent.
Two gates sit on top. OAuth clients ask for a read or write scope, and a read-scoped token that calls a mutating tool is refused before the tool runs. Role still decides what is possible: administrative tools require owner or admin, update_member_role is owner only and refuses to change your own role. API keys are the weak spot by design, and the docs say so plainly: a key is scoped to an organization and not to a role, so on the REST API anyone holding one gets that org's full read and write surface.
Pricing is quoted, and usage is metered on top
There is no pricing page, no price in the sitemap's 21 entries, and no figure anywhere on the site. The agency FAQ answers the question directly: pricing is scoped to your portfolio and call volume after a demo.
The shape of the bill is documented even though the numbers are not. The terms say fees are the plan you select plus per-minute usage charges above included volumes, billed in advance and non-refundable except where law requires. The Usage tab meters two kinds, conversations analyzed and transcript minutes, each against a monthly quota, any prepaid grant balance, and overage beyond both. Payments run through Stripe, with some invoices sent from Mercury, and custom agreements are charged as the agreement price plus usage above included amounts, first charge on the first of the following month.
Model spend is the variable nobody can size from outside. Filters, code stages and tool checks are free, AI analysis and flow checks are not, and the cost controls are real but manual: put cheap filters in front of expensive judgement, use exit() in a code stage to skip a later model call, and accept that a backfill of recent conversations spends per conversation and stops at 5,000.
The access trail is detailed, the audit is not finished
The audit log is unusually specific for a product at this stage. Reads of conversation content are recorded under a phi.* family named for the language a HIPAA business associate agreement asks for, and it applies to every organization whether or not the calls are healthcare calls: transcript reads, bulk queries with their row counts, exports, recording authorizations, and the nightly retention sweep. One row is written per request rather than per record. phi.cross_org_access records when Zelto staff read one of your conversations through internal support access, and staff rows carry a visible marker, so a customer can enumerate those reads. Rows are kept two years, the tab is read-only, and a member who navigates to it gets a 404.
What is missing is the attestation. The privacy policy claims only that the security program is aligned to SOC 2, and the trust center at trust.zelto.ai lists SOC 2 Type 2 as in progress against 25 policies and zero documents. Everything runs in AWS us-east-1 with every named sub-processor in the United States, so there is no residency choice; transfers out of Europe rest on Standard Contractual Clauses. Customer content is not used to train third parties' foundation models, though the terms' own license covers processing to improve the service. There is no uptime commitment: the terms disclaim that findings, transcriptions or suggested fixes will be accurate or complete.
Where it fits, and where it does not
The fit is a team already running production voice or chat agents on a platform Zelto ingests, with enough traffic that reading transcripts by hand has stopped working and enough engineering to act on a drafted prompt fix. The vendor claims 1M calls a month and more than 400 agents monitored across its customers, with three named logos and no independent review record to check that against. Agencies get the sharper version of the story: Companies attribute each conversation to a client, monitors can be scoped to companies rather than agents, and alerts route per client.
It is a poor fit in three cases. If you want to know the price before a call, you cannot. If you need a completed SOC 2 report, an uptime commitment or data kept outside the United States, none exists today. And if your stack is not one of the native integrations, someone has to write and maintain the ingest call, which is modest work but not no work.
Zelto price and terms
Terms
- Free plan
- YesYesThe custom model service has no published free plan and is sold through a demo
- Card to start
- YesNot required
- Sign up without sales
- PartlySome plans
The terms in full
The top banner offers the vendor's separate voice AI observability tool free: "Try our Langfuse for Voice AI for free". Its limits are not stated on this page.
For AI agents
- MCP server
- Yesdocs.zelto.ai,
- API
- Yesdocs.zelto.ai,
Where these prices come from
zelto.ai, checked
All plans and limits
| Plan | Price | Main limits |
|---|---|---|
| Zelto | Free | None stated |
Read from zelto.ai.
What we checked on Zelto
What the price counts
Priced only on a quote: zelto.ai/pricing redirects to the homepage, which shows no plans or figures and routes every buyer to scheduling a demo for a custom-trained model.
Custom models for voice AI agents. We train a model on your calls and your rules, benchmark every version, and run it behind your agent.
zelto.aiChecked
Free to use
The voice AI observability tool is offered free with a sign up; the page shows no free plan or price for the custom model training and hosting service itself.
Try our Langfuse for Voice AI for free
zelto.aiChecked
Public API
A documented REST API at api.zelto.ai covers agents, conversations, transcripts, findings and experiments, authenticated with an org-scoped bearer key.
Programmatic access to your Zelto organization. Every endpoint is scoped to the organization the API key belongs to. Mint a key at Settings →…
MCP server
A hosted MCP server gives AI clients read access to conversations, transcripts and findings plus a set of write tools, over an API key or OAuth sign-in.
Zelto exposes a Model Context Protocol server at https://api.zelto.ai/mcp. Any MCP-compatible client can connect and get read access to your agents,…
docs.zelto.ai/docs/mcpChecked
Open source
The platform is proprietary: the vendor claims ownership of the software and models and grants only a revocable right to use them during a subscription.
Zelto and its licensors own the Services, including all software, models, interfaces, and content we provide (excluding Customer Content), and all…
zelto.ai/termsChecked
Questions about Zelto
- How much does Zelto cost?
- Zelto is free to use. From zelto.ai, checked 18 November 2026.
- Does Zelto have an API or an MCP server?
- Zelto has an API. From docs.zelto.ai/api-reference/organizations/provision-an-organization-service-key, checked 20 October 2026. Zelto has an MCP server. From docs.zelto.ai/docs/mcp, checked 20 October 2026.
- Is Zelto open source, and can you host it yourself?
- Zelto is not open source. From zelto.ai/terms, checked 20 October 2026.
- What have we checked on Zelto?
- 5 facts, under these headings: What it costs to start, how you build with it.
- How do we check Zelto?
- We hold 5 facts on Zelto, quoted from 16 pages. Facts and price last checked 18 November 2026, when we also found the last change.
Sources and changes
Last changed
Sources16
- zelto.ai/privacy
- zelto.ai/terms
- zelto.ai/
- docs.zelto.ai/docs/monitors
- docs.zelto.ai/docs/monitors.md
- docs.zelto.ai/docs/mcp.md
- docs.zelto.ai/docs/quickstart
- docs.zelto.ai/docs/settings
- zelto.ai/for/voice-ai-agencies
- docs.zelto.ai/api-reference/organizations/provision-an-organization-service-key
- docs.zelto.ai/docs/integrations/webhooks
- docs.zelto.ai/docs/integrations/webhooks.md
- docs.zelto.ai/docs/mcp
- docs.zelto.ai/docs/settings.md
- zelto.ai/for/voice-ai-startups
- zelto.ai/integrations
What changed1
- What it does
Zelto price history
We have read the price once so far, on 18 November 2026: free.
| Price | Plan | First seen | Last seen | Readings | Source |
|---|---|---|---|---|---|
| Free | 18 November 2026 | 18 November 2026 | 1 | Pricing page, read 18 November 2026 |
From our research on 18 November 2026. A date is when we read the page, not when the price changed.
Similar in Agent tools and integrations
See all in Agent tools and integrations
- ERIUS PHONERented cloud Android phones, each in its own container, that keep an agent's apps and sign-ins between sessions.
- TinyFishWeb search, page fetching, cloud browsers and goal-driven web automation for AI agents, with search and fetch free.
- SmitheryPublic MCP server registry, now run by Arcade, listing direct vendor MCP URLs after its proxy shut down.
- UI DiscoveryA library of real websites, iOS apps and dashboards with each one's palette, fonts and original build prompt.
Claimed by the Zelto team.
We check its facts against the Zelto site.Facts checked 18 November 2026
Report an error if a fact is wrong.