Starter
$500/month
$6,000 billed yearly
- 10 subjects
- 50 sources
- 20 signals
- 7M LLM tokens / month
Feed and source watching, evidence capture, pulses, and LLM scoring. The whole product — no tier removes part of it.
Get startedBrand & subject intelligence
Every signal, traced back to
the thing that caused it.
SignalReact watches the places your brand actually lives — feeds, sites, individual pages — scores what turns up against signals you define, and keeps the whole chain of reasoning behind every score: the artifact it read, the prompt version, the model that answered, and how the vote went.
- Sources watched per tenant
- 50 — 500+
- Scoring provenance kept
- 100%
- Billed overage
- None — ever
Monitoring tools are loud. That is the problem.
A keyword alert fires on volume. It cannot tell you whether a mention mattered, it cannot explain why it scored the way it did, and it certainly cannot be handed to someone who needs to act on it.
SignalReact is built the other way round. It stays quiet until something you defined as meaningful actually happens — and when it speaks, the evidence comes with it.
How it works
Five steps, and you can inspect every one.
-
01
Define the subject
A brand, a company, a person, or a topic. You attach the sources it appears in and the signals you care about — each with its own weight and its own evaluation instructions.
-
02
Capture the evidence
Every artifact is stored once. Identity resolves by provider ID first, then canonical URL, then content hash — so the same article arriving through three routes stays one record instead of three duplicates.
-
03
Evaluate
Rules short-circuit what they can answer cheaply. Everything else goes to a model, optionally to several for consensus. Every constituent call is recorded, including the retries.
-
04
Emit a pulse
One pulse per artifact, subject, and signal. Scores roll into a subject score with recency decay, so an old crisis stops dominating a current picture.
-
05
React
Signed webhooks or SQS, delivered per watch rather than per pulse — two teams watching the same subject each get their own auditable, independently retryable delivery.
Provenance
“Why did it score 0.82?” has an answer.
Most tools give you a number. When someone senior asks where it came from, there is nothing behind it. SignalReact treats that question as a first-class requirement.
Every evaluation is immutable and keeps one attempt record per underlying call — the real provider and model version that answered, the prompt version, the rule version, token counts, cost, latency, and a digest of the raw response. The selected answer and the reason it was selected are stored alongside them.
Nothing is overwritten. Re-scoring a subject adds provenance; it never replaces it.
- Decision
- consensus · 2 of 3
- Selected by
- aggregator attempt
- Prompt version
- v4
- Rule version
- v2 — no short-circuit
- Attempts recorded
- 4 (1 retry)
- Tokens
- 11,204
- Evidence
- sha256:9f3c…a7e1
Illustrative shape. Every field shown is stored per evaluation.
Page & design watching
Watch a page the way a person would.
Feeds only cover what publishes a feed. A competitor's pricing page, a partner's terms, a regulator's guidance page — none of them announce a change.
SignalReact renders those pages in a real browser, fingerprints what it sees, and emits a pulse only when the page genuinely moves. A first sighting records a baseline and stays silent, rather than inventing a change out of having nothing to compare against.
Included on Enterprise
competitor.com/pricing
baseline recorded
competitor.com/pricing
no change
competitor.com/pricing
changed · pulse emitted
Pricing
Three plans. No billed overage on any of them.
Allowances are ceilings, not soft targets. You are warned at 70%, warned again at 90%, and work stops at 100% — so nobody receives a surprise invoice.
Most chosen
Growth
$1,500/month
$18,000 billed yearly
- 35 subjects
- 175 sources
- 70 signals
- 20M LLM tokens / month
- Webhook and SQS reactions
Adds the ability to drive systems you already run, so a signal becomes an action instead of a notification.
Get startedEnterprise
from$3,500/month
from $42,000 billed yearly
- 100+ subjects
- 500+ sources
- 200+ signals
- 50M+ LLM tokens / month
- Page and design watching
- Bring your own Bedrock
- SSO, custom retention, dedicated support
Bring-your-own Bedrock moves inference into your own AWS account, which removes the only variable cost from the contract.
Talk to usA “+” marks a floor that is negotiated upward, not an unmetered allowance — every plan is enforced by the same meters.
Start watching something today.
Create a workspace, point it at a subject, and see the first pulse arrive with its evidence attached.