RXed AI News

AI to the bone.
@RXed_EU
Audited 2026-08-16 · RXed table v1.0

Cekura

Visit cekura.ai
“Automated QA for Voice AI and Chat AI Agents” — the vendor’s own words

The lowest-friction way into voice-agent QA — $0.25 a test minute, no sales call, an MCP server and Claude Skills on every plan — but it grades transcripts rather than audio, its testing is state-blind, and most of the public evidence for how good its evaluations are was written by Cekura.

Best for: Small teams shipping a voice or chat agent who want structured simulation and CI gating running this week for tens of euros, and who will calibrate the scores themselves before trusting them.
Scope18/20
Quality6/10
Where the quality sits
6Reactive
5Retrieval & Memory
6Orchestration
7Validation
5Models
SpecialistAutomation & AgentsVoice & SpeechSecurity & ComplianceFreemium
Vendor
Cekura (Tatva Labs Inc.) · www.cekura.ai
Origin
US — San Francisco
Pricing
Pay as you go $0 to start · Startup $500/mo · Enterprise Custom
Users (official only)
75+ customers across healthcare, BFSI, logistics, recruitment and retail (source, 2025-07-03)
Pay as you go$0 to start$0.25/voice testing minute, $0.025/chat reply, $0.05/monitored call; 10 concurrent calls; 1 seat free, $30/mo per extra seat; 30-day retention
Startup$500/mo10,000 credits ≈ 2,000 test minutes; 50 concurrent calls; 10 seats; signed BAA & DPA; 90-day retention
EnterpriseCustomAnnual; red teaming, load testing, vendor bake-offs, VPC/on-prem, SSO/SCIM, audit logs, forward-deployed engineers

The pricing model changed during 2026. Coverage from February to June 2026 describes a credit-based Developer plan at $30/user/month (750 credits, roughly 5 credits per test minute); the live page on 16/08/2026 is per-minute pay-as-you-go with a $500 Startup tier. HIPAA and GDPR are now listed as included on every plan, where a March 2026 comparison described them as Enterprise-only. Re-check before budgeting.

checked 2026-08-16 · vendor pricing page

Element scores

Reactive
Retrieval & Memory
Orchestration
Validation
Models
Primitives
Pr7
Prompts
Em4
Embeddings
Cx6
Context
Tr7
Tracing
Lg5
LLM
Compositions
Fc7
Function calling
Vx
Vector store
Rg4
RAG
Gr7
Guardrails
Mm6
Multimodal
Deployment
Ag7
Agents
Ft4
Fine-tuning
Fw7
Frameworks & harnesses
Ev8
Evaluations
Sm4
Small models
Emerging
Ma5
Multi-agent
Sy8
Synthetic data
Pc7
Protocols
In7
Interpretability
Th
Thinking models
Tap or hover any element to see why it got that score.

Strengths

The cheapest credible entry point in voice-agent QA: $0.25 a test minute, 300 free credits, no card and no sales call, with an MCP server and Claude Skills on every plan and unlimited Python metrics free. Coverage is wide for a company founded in 2024 — simulation, red teaming across jailbreak, bias, toxicity and PII leakage, production call monitoring, cross-vendor bake-offs, OpenTelemetry spans across LLM, TTS, STT and tool calls, and Insights clustering failing evals into a ranked list of root causes rather than a pile of logs. Integrations cover Retell, Vapi, ElevenLabs, LiveKit, Pipecat, SIP, Twilio and Plivo.

Honest dings

Evaluation is transcript-primary, so the audio failures that make a voice agent feel wrong need explicit configuration to catch. Testing is state-blind — it cannot set a pre-condition or verify a post-call state change — and there is no CLI. The one cross-platform human benchmark (arXiv 2511.04133, Evalion-co-authored) puts the non-leading platforms at 0.73 F1 against the leader's 0.92. A May 2026 independent roundup notes that a large share of the public evidence base, including the compliance claims, is Cekura-authored. Pricing changed shape twice in six months.

Prices and details change — this passport is re-verified at least quarterly.
Sources (10) — every claim traceable

Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).

  • cekura.ai/pricing — Live pricing 16/08/2026: $0.25/voice minute, $0.025/chat reply, $0.05/monitored call, Startup $500/mo, seat and retention limits; MCP server, Claude Skills, HIPAA and GDPR listed on every plan (accessed 2026-08-16)
  • docs.cekura.ai/llms.txt — Official docs index: LLM Judge and Python metrics, rubrics, Conditional Actions, Mock Tools, dynamic variables, test profiles, load and infrastructure testing, integrations (Retell, Vapi, ElevenLabs, LiveKit, Pipecat, SIP, Twilio, Plivo) (accessed 2026-08-16)
  • ycombinator.com/companies/cekura-ai — YC Fall 2024, three IIT Bombay founders, 75+ customers claim, and the Red Teaming launch post detailing jailbreak, bias, toxicity and PII-leakage categories plus Red Teaming as a Service (accessed 2026-08-16)
  • economictimes.indiatimes.com/tech/startups/yc-backe… — Independent reporting of the $2.4M seed and cofounder Sidhant Kabra's on-record 75+ customer figure and Bangalore office plan (03/07/2025) (accessed 2026-08-16)
  • businessinsider.com/read-pitch-deck-y-combinator-st… — Independent reporting of the seed round, the March 2025 rebrand from Vocera, founder backgrounds and team size at the time (seven employees) (accessed 2026-08-16)
  • speechmatics.com/company/articles-and-news/de-risk-… — Independent 11-platform roundup (26/05/2026): flags Cekura's SOC 2 / HIPAA / GDPR claims as vendor-authored, notes a large share of the public evidence base is Cekura-written, and records the then-current credit pricing and integration list (accessed 2026-08-16)
  • arxiv.org/abs/2511.04133 — Testing the Testers (Andrés et al., 06/11/2025) — the only published cross-platform human benchmark of voice AI testing platforms, 21,600 human judgments; top platform 0.92 F1 evaluation quality against 0.73 for the others. Co-authored by Evalion, a competitor, so read the ranking with that in mind (accessed 2026-08-16)
  • coval.ai/blog/coval-vs-cekura — Competitor-authored comparison (Coval, 17/03/2026). Used only for the points it concedes against its own interest — Cekura was first in the category to ship an MCP server — and for the dated state-blind and no-CLI observations (accessed 2026-08-16)
  • linkedin.com/company/cekuraai — Official June 2026 product update: OpenTelemetry spans for LLM, TTS, STT and tool calls, Insights root-cause clustering, Cekura Academy, Custom Skills for the Cekura Agent (accessed 2026-08-16)
  • sigmamind.ai/blog/ai-voice-agent-testing-platform-b… — Independent buyer's guide: 30+ language simulation, frequency-based load testing and the 10-concurrent-call developer cap, and the observation that Cekura is transcript-centric by default (accessed 2026-08-16)