Signal Run
Start free
[PULSE AI]
AVAILABLE

Ship governed AI agents on your own data.

A gateway, guardrails, evals, and an audit trail — billed by use.

trace · run_4e2a · agent: support-triage2.18s · 7 spans · $0.027
guardrail.PII_REDACT
redacted 2 spans (email, card)14ms
tool.ontology.Account.read
Account{plan=team}38ms
llm.claude-sonnet-4.6
billing topic · normal priority1,284ms
tool.kb.search
Chunk[3] · refund_policy184ms
guardrail.NO_PROMISES
held · no refund committed18ms
eval.faithfulness
pass 0.9838ms
audit.write
immutable · 0x4e2a…91d9ms
FIG 00 · TRACE · GUARDRAIL → TOOL → MODEL → GUARDRAIL · ILLUSTRATIVE
01HOW IT WORKS

The agent lifecycle, governed end to end.

01

Connect data securely

Query your warehouse in place; per-tenant isolation from the first call.

02

Build in the sandbox

Pick a model through one gateway, add tools and knowledge, test in the Playground.

03

Govern

Attach guardrails and an eval set. Gate releases on pass-rate.

04

Deploy through the gateway

One governed endpoint, six surfaces. Guardrails, audit, and billing on every call.

02PLAYGROUND

Build and test in a sandbox before anything ships.

Pulse AIagent support-triage
PlaygroundPromptToolsKnowledgeGuardrailsEvalsTracesDeploy
sandbox MC
Agents7
Routines6
Tools18
Models7
Knowledge6
Guardrails14
Evals42
Traces1.4M
Sandbox · build and test before you shipclaude-sonnet-4.6
U
A customer wrote: 'I was charged twice this month, please fix it.' What queue?
Classified → queue: billing · priority: high · sentiment: frustrated. Policy NO_PROMISES held (no refund committed). Suggested first reply drafted.
governedtrace_4e2a1,204 in · 312 out$0.0091840 ms
U
And this one: 'How do I export my data to CSV?'
Classified → queue: support · priority: normal. Grounded answer found in product_docs (3 chunks). Confidence 0.96.
governedtrace_51c8980 in · 240 out$0.0071520 ms
Message the agent…Run
[MODEL]
Claude Sonnet 4.6
anthropic1.4 s p99
in $3.00out $15.00 /1M
Claude Opus 4.7
anthropic2.8 s p99
in $18.00out $72.00 /1M
Claude Haiku 3.6
anthropic0.4 s p99
in $0.80out $2.40 /1M
GPT-5.2
openai1.8 s p99
in $5.00out $20.00 /1M
[LAST RUN]
tokens in4,218
tokens out982
cost$0.027
latency2.18 s
eval96.4%
FIG 02.1 · PULSE AI PLAYGROUND · TOKENS / COST / LATENCY PER RUN · ILLUSTRATIVE
03ROUTINES

Six patterns to start from.

Each routine maps a trigger to an agent to a deploy target. Start from one and customize.

[ROUTINE]

Scheduled routine

A recurring, multi-step agent that runs on a schedule and reports back — daily digests, triage sweeps, recurring reviews.

schedule · cronScheduled / batch
Start with this routine
SANDBOXED
[COMPUTER-USE]

Browser automation

An agent that drives a real browser through a sandboxed computer-use runtime — fill forms, gather data, reconcile portals.

manual · APIREST API + sandbox
Start with this routine
[EVENT-DRIVEN]

Inbox / event agent

Fires on an incoming event — a new email, ticket, row, or upload — runs the agent, and routes or replies.

webhook · eventWebhook
Start with this routine
[EXTRACTION]

Document pipeline

Drop files in, get structured objects out. Chunk, extract, validate against a schema, and land the result.

upload · batchWebhook + batch
Start with this routine
[ASSISTANT]

Embedded copilot

A grounded chat assistant embedded in your own product via the widget or API — answers from your KB, with citations.

user messageEmbed widget / API
Start with this routine
ADVANCED
[ORCHESTRATION]

Multi-agent chain

One agent calls another as a tool — a router hands off to specialists. Each hop is governed and traced.

API · agent-to-agentAPI + MCP
Start with this routine
04DEPLOY

One governed gateway. Six surfaces.

Deploy widens reach, never permission — guardrails, audit, and billing apply to every call.

Governed gateway
identify · authorize · throttle · meter · govern · observe
ACTIVE
REST API + SDK
POST /v1/agents/support-triage/run
ACTIVE
Webhook
on event → run agent
ACTIVE
Sandbox / Chat
in-platform Playground
SOON
Embed widget
script snippet
SOON
Slack / Teams
chat-ops connector
SOON
Scheduled
cron over a dataset
[API + KEYS]
POST https://api.signalrun.io/v1/agents/support-triage/run Authorization: Bearer sk_live_8f2a… Content-Type: application/json { "input": "Route ticket#T-44190" }

One host. Your tenant is resolved from the key — isolated by default.

sk_live_8f2a…productionagents:run
sk_test_3c91…sandboxagents:run
Per-env, scoped, hashed, shown once, rotatable, fail-closed. Vanity subdomains, IP allowlists & VPC are Enterprise.
[API MANAGEMENT, BUILT IN]Every API call runs the same guardrails, audit, and metering as the in-app sandbox — multi-tenant, isolated by default, metered per call.
05GOVERNANCE

The part most agent stacks skip.

Guardrails

Input and output policies — PII redaction, no-promises, escalation — enforced before and after every call.

Evals

Datasets and pass-rate gating. Catch regressions before they ship; gate releases on a threshold.

Audit-immutable log

Every run, guardrail decision, and write-back carries actor, intent, and a policy decision id — append-only.

Traces

Span-level visibility into every model call, tool, and guardrail. Export and retain per plan.

06MODELS

One gateway, many models.

Route by cost, latency, or region. Bring your own keys; they decrypt only inside the gateway runtime.

MODELPROVIDERIN / 1MOUT / 1MP99REGION
Claude Sonnet 4.6anthropic$3.00$15.001.4 sUS-East
Claude Opus 4.7anthropic$18.00$72.002.8 sUS-East
Claude Haiku 3.6anthropic$0.80$2.400.4 sUS-East
GPT-5.2openai$5.00$20.001.8 sUS-East
Gemini 2.5 Progoogle$2.50$10.001.6 sEU-West
Llama 4.1 70Bmeta$0.40$0.800.9 sOn-prem
Mistral Large 3mistral$2.00$6.001.2 sEU-West
DeepSeek V3deepseek$0.50$2.001.3 sOn-prem
FIG 06.1 · GATEWAY MODEL CATALOG · ILLUSTRATIVE PRICING
[START FREE]

Land, try, and pay — without a human.

Free to start: 1 agent, 1k runs/mo, no card. Per-seat plus metered usage when you grow.

Start freeCompare plans