Signal vs Bluejay
Bluejay simulates callers and groups metric failures into issues. Signal names the step, the cause and the change, from your Retell calls.
In short
Bluejay (YC 2025) tests voice agents with simulated callers it calls Digital Humans, monitors production calls against metrics you define, groups failures into Issues, and offers Self-Improve, which rewrites the prompt until a test suite passes. Signal reads your Retell agent's flow and production calls, places each repeated failure on a step with a cause and a suggested change, and tells you whether a new version made things better for real callers. Bluejay's issue is 'these calls failed the same metric'. Signal's is 'this step, this cause, this change'.
Side by side
- Main job
- BluejaySimulate (Digital Humans) and monitor
- SignalFind what to fix in production and check that the fix worked
- Started from
- BluejayPre-launch testing
- SignalUnderstanding production calls
- Retell connection
- BluejayPaste their URL into Retell's agent webhook field, then sync agents
- SignalAPI key; reads config and calls over the API; no webhook
- Agent config read
- BluejayAgent list, prompt versions
- SignalFull flow: steps, transitions, tools, versions
- A call is graded against
- BluejayCustom metrics: pass or fail, number, text, enum, JSON, tool call; 70+ in total
- SignalA one-page brief you approve; outcomes from platform facts first
- An issue is
- BluejayGrouped metric failures, tagged pattern or error; grouping not editable
- SignalOne change that fixes a group of calls, with place and cause; you can reject it
- Flow step in the call model
- BluejayNo field for a step or node
- SignalEvery call is a trace through your flow's steps
- Configured flow vs what production did
- BluejayNo
- SignalYes, your flow with production traffic on it
- Version comparison
- BluejayVersion labels; 'production' label moves only with a human
- SignalBefore and after on production calls, with a verdict only when the difference is real
- Fix loop
- BluejaySelf-Improve rewrites the prompt only, graded on simulations
- SignalSuggested fix per issue; Copilot drafts a change and runs simulated tests; you publish
- Simulation
- BluejayYes: Digital Humans, 500+ variables
- SignalReplays of your stored production calls only
- Compliance
- BluejaySOC 2 Type II; BAA from Growth
- SignalNone. PII handled, no PHI
- How they bill
- BluejayPay as you go, or monthly plans with a pool of simulation and monitoring minutes
- SignalFlat monthly plans with one credit balance; no charge per metric; no overage
Choose Bluejay if
- You need simulated callers before launch, with control over voice, accent, language, emotion and noise (Bluejay lists 500+ variables).
- You want red teaming or uptime checks that place real test calls on a schedule.
- You need SOC 2 Type II, or a BAA (from their Growth plan).
- You want an automatic loop that rewrites your prompt and pushes it to Retell, Vapi or ElevenLabs until tests pass.
- You want deterministic tool-call metrics alongside LLM-judged ones, with a Not Applicable result.
Choose Signal if
- You want each issue placed on the step, tool or transition where it happens, with a cause and a suggested change.
- You want to say an issue is wrong and have that stick on the next run.
- You want to keep your Retell webhook pointed at your own backend.
- You want to approve a brief rather than define metrics.
- You want 'the fix worked' measured on production calls, with a verdict only when the difference is real.
What the same calls cost across the market
Every vendor at its list price for the same customer at 1,000, 10,000 and 50,000 calls a month: 3-minute calls, one seat, monitoring only, no simulation.
| Vendor | 1,000 calls | 10,000 calls | 50,000 calls |
|---|---|---|---|
| Signal, monthly | $49Starter, $0.049 a call | $499Scale, $0.050 a call | from $1,500Enterprise, $0.030 a call |
| Signal, annual | $39Starter, $0.039 a call | $425Scale, $0.043 a call | CustomEnterprise |
| Tuner | $39Builder, $0.039 a call | $376.50Builder plus overage, $0.038 a call | $1,729Growth plus overage, $0.035 a call |
| Cekura | $50Pay as you go, $0.05 a call | $500$0.05 a call | $2,500$0.05 a call |
| Coval | $100Starter, $0.10 a call | $500Growth, $0.05 a call | $2,500 to $6,500$0.05 to $0.13 a call |
| Bluejay, the vendor this page compares | $500Growth, $0.50 a call | $1,000Scale, $0.10 a call | about $6,100$0.12 a call, derived |
| Roark | $500Team, $0.50 a call | $3,720Team, $0.37 a call | $18,600Team, $0.37 a call |
| Retell AI QA add-on | $290$0.29 a call | $2,990$0.30 a call | $14,990$0.30 a call |
| Hamming | Contact sales only | Contact sales only | Contact sales only |
List prices read on from each vendor's pricing page.
Signal's price includes the copilot, simulated tests, the API and the MCP server, all paid from its credits. The other figures cover monitoring only.
How each figure was built
- Signal
- Computed from Signal's plans: the smallest plan with enough credits for the volume at 3 minutes a call, at its monthly or annual price.
- Tuner
- About 5 credits a call (from "5,000 credits (~1,000 calls)"). 10,000 calls is 50,000 credits, $39 plus 45,000 × $0.0075. 50,000 calls is 250,000 credits, $399 plus 190,000 × $0.007. Tuner shows no annual discount. Tuner pricing
- Cekura
- "$0.05 per monitored call", flat, so length does not matter. Cekura pricing
- Coval
- Growth $500 plus 40,000 calls over. The page prints the overage as "$0.05/monitoring min" while including calls, so the unit is ambiguous: $2,000 over if per call, $6,000 over if per minute. Coval pricing
- Bluejay
- 150,000 monitoring minutes against 34,000 included, 116,000 over at 1.5x of the implied $1,000 / 34,000 a minute. The implied rate is derived, not printed. Bluejay pricing
- Roark
- 5 metrics at $0.02 each plus $0.024 transcription is $0.124 a minute. At 1,000 calls usage is $372, so the $500 minimum applies. At 50,000 calls Roark would negotiate an Enterprise deal (floor "$4,000+"). Roark pricing
- Retell AI QA add-on
- "$0.10/min" after "First 100 minutes free", assuming the free minutes renew monthly (the page does not say). It sits on top of Retell's voice fees. Retell pricing
- Hamming
- Hamming prints no price. Its own blog estimates "~$2-5K/month for most teams", a vendor estimate and not a price. Hamming pricing Hamming blog estimate
How Bluejay works
Bluejay’s pitch is “simulate a month’s worth of customer interaction in 5 minutes”. It has three parts: Test, Monitor, Improve.
Test. Digital Humans (marketed as Mimic) are synthetic callers with more than 500 variables: voice, accent, language, emotion, background noise. There is also red teaming with an attack catalogue, uptime monitoring that places real test calls on a schedule, and a benchmark.
Monitor. For Retell, you store an API key, paste Bluejay’s URL into Retell’s agent-level webhook field, sync your agents and link the agent id. Every call after that streams in. Their docs do not describe a backfill. Calls are scored by custom metrics you define: pass or fail, number, text, enum, JSON, and a deterministic tool-call metric that uses no model.
Issues. Failures on evaluated production calls become Issues. Bluejay groups them automatically, and you cannot edit the grouping. Each issue has a title, its calls, a status, a priority and a tag: pattern or error. Their call model has no field for a flow step or node; a call is a transcript, tool calls and free-form events.
Self-Improve. Started from Bluejay’s chat. Each round freezes the pass or fail metrics, runs a baseline, picks one failure pattern, rewrites the prompt, pushes it to your platform and runs again. A better result is kept as a new version; the production label only moves when a person moves it. It changes the prompt only, never tools or flow.
Where Signal differs
The issue. Bluejay’s Issues are the closest thing to Signal‘s board, and they show the gap. A Bluejay issue is a set of calls that failed the same metric. A Signal issue is a set of calls that one change would fix, placed on the step where it happens, with the cause and the evidence. If it is wrong, you reject it with a reason and it stays rejected.
Your webhook. Bluejay asks you to point Retell’s agent webhook at them. Signal reads over the API and leaves your webhook alone.
Nothing to define. Bluejay grades against metrics you or it define. Signal drafts a one-page brief from your configuration and you approve it.
Proof. Self-Improve is graded on simulations. Signal measures a new version on production calls and says whether the difference is real, or that there are not enough calls yet.
Where Bluejay is stronger. Simulated callers with fine control, red teaming, scheduled uptime calls, several platforms, compliance (SOC 2 Type II, BAA from Growth), and an automatic prompt loop. Signal does none of these.
What it costs
Bluejay offers pay as you go with a starting credit, and monthly plans that include a pool of minutes. The pool looks shared between simulation and production monitoring rather than two separate allowances, and past it minutes burn at a higher rate. A BAA comes with their Growth plan, load testing with Scale, and SSO with Enterprise.
The price table on this page compares both at 10,000 three-minute calls a month. Bluejay’s price also covers simulation, which Signal does not sell.
Signal charges a flat monthly plan with one balance of credits: each minute of a call costs a credit once it has started, and tests and the copilot draw on the same balance. Adding a check does not change the bill, and there is no charge per metric. There are no overage charges either: calls past the balance are stored, and analysed when the month resets or after an upgrade.
Using both
There is one thing to plan for: Bluejay takes the Retell agent webhook, and Signal does not need it, so they do not conflict.
- Signal names the issue on production: the step, the cause, the calls.
- You build a Bluejay simulation or metric aimed at exactly that issue.
- Bluejay’s Digital Humans or Self-Improve work on it before release.
- After release, Signal checks production: did the issue shrink, and did anything else get worse.
Checked on 22 September 2026 against Bluejay's public pages.getbluejay.ai