00VANTAGE · BOOTING CONTROL PLANE

Agent control plane

Ship agents you can actually watch.

Vantage gives every agent run a plan, a trace, and a verdict. Deploy, observe, and roll back from one console, without wiring your own infrastructure.

Start free

Read the docs

No card required · SOC 2 Type II · Self-host available

run_9f2c1a · billing-triage-agentPassed

Received0.2s

Planned1.4s

Tools called6 calls

Verified0.8s

Delivered12.3s total

run_9f2c1a · billing-triage-agentPassed

Received0.2s

Planned1.4s

Tools called6 calls

Verified0.8s

Delivered12.3s total

--:--:-- UTC

Running in production at

The gap

Most teams can build an agent in a week and cannot explain it in a quarter.

Prototypes pass demos. Production runs fail quietly, at 3am, on the one input nobody tested. Vantage is for everything after the demo.

of agent teams ship without run-level tracing

median time to diagnose a failed run by hand

production runs silently returns a wrong result

The loop

Plan, run, verify. Every time, on the record.

01 Plan

Every request becomes a written plan.

Before a single tool fires, the agent commits to a plan you can read, diff, and approve. No guessing at intent after the fact.

plan.md

DRAFT

Fetch invoice INV-40912 from billing

Classify dispute reason from thread

Check refund policy for tier: Team

Draft reply, hold for approval

sandbox · scoped

6 tools

billing.read

mail.draft

timeout 30s

billing.refund ≤ $200

rate 4/min

crm.read

02 Run

Tool calls execute inside a sandbox you control.

Scoped credentials, per-tool rate limits, and a hard timeout on every step. Runs stay inside the blast radius you set.

03 Verify

Nothing ships without passing a check.

Attach evals, assertions, or a human approver to any run. Failed verdicts roll back automatically and page the owner.

verify · strict

4 / 4

policy_match

PASS

amount_within_cap

PASS

no_pii_in_reply

PASS

human_approved

PASS

Capabilities

Everything production asks for on day one.

Run traces

Step-level timing, inputs, outputs, and token cost for every run.

Run traces

Step-level timing, inputs, outputs, and token cost for every run.

Cost controls

Per-agent, per-team, and per-customer spend caps.

Cost controls

Per-agent, per-team, and per-customer spend caps.

Evals

Run your test suite against live traffic, not just fixtures.

Evals

Run your test suite against live traffic, not just fixtures.

Human in the loop

Route any step to a reviewer queue with full context attached.

Human in the loop

Route any step to a reviewer queue with full context attached.

Rollbacks

Pin a known-good version and revert in one click.

Rollbacks

Pin a known-good version and revert in one click.

Model routing

Swap providers per task without touching agent code.

Model routing

Swap providers per task without touching agent code.

The console

One screen for every agent you run.

Filter by agent, version, customer, or verdict. Open any run and read it top to bottom, exactly as it happened.

Run	Agent	Duration	Cost	Verdict
run_9f2c1a	billing-triage	12.3s	$0.041	Passed
run_9f2b84	invoice-extract	4.8s	$0.012	Passed
run_9f2b12	refund-approver	31.6s	$0.118	Review
run_9f2a77	billing-triage	9.9s	$0.037	Passed
run_9f2a05	contract-parse	2.1s	$0.008	Failed

run_9f2c1aPassed

billing-triage12.3s · $0.041

run_9f2b84Passed

invoice-extract4.8s · $0.012

run_9f2b12Review

refund-approver31.6s · $0.118

run_9f2a77Passed

billing-triage9.9s · $0.037

run_9f2a05Failed

contract-parse2.1s · $0.008

curl https://api.vantage.dev/v1/runs \
  -H "Authorization: Bearer $VANTAGE_KEY" \
  -d agent=billing-triage \
  -d verify=strict

Integrations

Keep your stack. Add the record.

Connect the tools your agents already touch. Vantage reads what they know and writes back every verdict, with no glue code in between.

Observability

Beacon · Meridian

Observability

Beacon · Meridian

Data

Ledgerly · Holloway

Data

Ledgerly · Holloway

Integrations

Native integrations

Plugs into the stack you already run.

Ship traces to the tools you already pay for. Pull context from the systems your agents act on.

Beacon
Ledgerly
Cortex
Keyring
Relaypoint
Casefile
Meridian
Holloway
Nearfield
Pagewire

Vector

Cortex · Nearfield

Identity

Keyring

Messaging

Relaypoint · Pagewire

Messaging

Relaypoint · Pagewire

Ticketing

Casefile

Bring your own

Webhooks · OpenTelemetry

Trust

Built for teams that get audited.

Compliance

SOC 2 Type II, ISO 27001, HIPAA on request

Compliance

SOC 2 Type II, ISO 27001, HIPAA on request

Data residency

US, EU, and self-hosted

Data residency

US, EU, and self-hosted

Retention

Configurable, 0 to 400 days

Retention

Configurable, 0 to 400 days

Encryption

AES-256 at rest, TLS 1.3 in transit

Encryption

AES-256 at rest, TLS 1.3 in transit

Access

SSO, SCIM, and role-based permissions

Access

SSO, SCIM, and role-based permissions

Training

Your data is never used to train models

Training

Your data is never used to train models

SOC 2 Type II

ISO 27001

GDPR

Read the security overview

Evidence

Teams stopped guessing.

We cut incident triage from an afternoon to about six minutes. The trace tells you where it went wrong before you finish reading the alert.
PR
Priya Raghunathan
Staff Engineer, Northwind
We cut incident triage from an afternoon to about six minutes. The trace tells you where it went wrong before you finish reading the alert.
PR
Priya Raghunathan
Staff Engineer, Northwind
The approval queue is what got us through legal review. Every high-risk action now has a name attached to it.
DO
Daniel Okafor
Head of Platform, Lattice Health
The approval queue is what got us through legal review. Every high-risk action now has a name attached to it.
DO
Daniel Okafor
Head of Platform, Lattice Health
Evals run against live traffic now, not a frozen test set. We catch regressions the morning they ship, not a week later.
MT
Mei Tanaka
ML Lead, Fieldstone
Evals run against live traffic now, not a frozen test set. We catch regressions the morning they ship, not a week later.
MT
Mei Tanaka
ML Lead, Fieldstone
Per-customer spend caps ended our surprise invoices. Finance stopped asking what the agents were doing with the budget.
LM
Lucas Moreau
VP Engineering, Brightwell
Per-customer spend caps ended our surprise invoices. Finance stopped asking what the agents were doing with the budget.
LM
Lucas Moreau
VP Engineering, Brightwell
Rolling back an agent used to mean a redeploy. Now it is one click, and the run history shows exactly what changed.
AN
Amara Nwosu
SRE Manager, Corvid Pay
Rolling back an agent used to mean a redeploy. Now it is one click, and the run history shows exactly what changed.
AN
Amara Nwosu
SRE Manager, Corvid Pay

of runs verified automatically

median time to triage a failed run

more agents shipped per quarter

Changelog

Shipping every week, on the record too.

Every release is versioned, signed, and rolled out behind a flag. Here is what changed lately.

Oct 2, 2026

v2.14

Replay any run on a new model

Re-execute a past trace step by step on a different provider and diff the verdicts.

Oct 2, 2026

v2.14

Replay any run on a new model

Re-execute a past trace step by step on a different provider and diff the verdicts.

Sep 18, 2026

v2.13

Approval queues in Slack

Approve or reject high-risk steps without leaving the thread. Every decision lands on the trace.

Sep 18, 2026

v2.13

Approval queues in Slack

Approve or reject high-risk steps without leaving the thread. Every decision lands on the trace.

Sep 4, 2026

v2.12

Spend caps per customer

Hard limits per tenant. Runs that would cross the cap pause and page the owner.

Sep 4, 2026

v2.12

Spend caps per customer

Hard limits per tenant. Runs that would cross the cap pause and page the owner.

Aug 21, 2026

v2.11

EU data residency

Traces, evals, and artifacts can now stay in Frankfurt end to end.

Aug 21, 2026

v2.11

EU data residency

Traces, evals, and artifacts can now stay in Frankfurt end to end.

Plan
Run
Verify
Roll back
On the record

Pricing

Priced per run, not per seat.

Invite the whole team on any plan. You pay for what your agents actually do.

FREE

No card required

For a first agent in staging.

1,000 runs per month

1 seat

7-day trace retention

Run traces and step timing

Community support

Start free

Monthly

Annual

Save 20%

TEAM

/ month

Billed monthly

For teams running agents in production.

50,000 runs per month

Unlimited seats

90-day retention

Evals and assertions

Approval queues

Email support

Rollbacks and version pins

Start 14-day trial

ENTERPRISE

Custom

Annual contract

For regulated and high-volume workloads.

Unlimited runs

Self-host or dedicated VPC

400-day retention

SCIM and audit log export

Named support engineer

99.9% SLA

Talk to sales

FREE

No card required

For a first agent in staging.

1,000 runs per month

1 seat

7-day trace retention

Run traces and step timing

Community support

Start free

Monthly

Annual

Save 20%

TEAM

/ month

Billed monthly

For teams running agents in production.

50,000 runs per month

Unlimited seats

90-day retention

Evals and assertions

Approval queues

Email support

Rollbacks and version pins

Start 14-day trial

ENTERPRISE

Custom

Annual contract

For regulated and high-volume workloads.

Unlimited runs

Self-host or dedicated VPC

400-day retention

SCIM and audit log export

Named support engineer

99.9% SLA

Talk to sales

FREE

No card required

For a first agent in staging.

1,000 runs per month

1 seat

7-day trace retention

Run traces and step timing

Community support

Start free

Monthly

Annual

Save 20%

TEAM

/ month

Billed monthly

For teams running agents in production.

50,000 runs per month

Unlimited seats

90-day retention

Evals and assertions

Approval queues

Email support

Rollbacks and version pins

Start 14-day trial

ENTERPRISE

Custom

Annual contract

For regulated and high-volume workloads.

Unlimited runs

Self-host or dedicated VPC

400-day retention

SCIM and audit log export

Named support engineer

99.9% SLA

Talk to sales

Additional runs at $0.004 each. Failed runs are never charged.

Questions

Before you ask.

Do I have to rewrite my agents to use Vantage?

No. Wrap your existing calls with the SDK, or point your gateway at the Vantage endpoint. Most teams are sending traces within an afternoon.

Which models and frameworks are supported?

Any provider with an HTTP API, plus the common orchestration frameworks out of the box. Model routing is configuration, not code.

Can we self-host?

Yes. The full control plane runs in your own VPC on the Enterprise plan, with the same console and SDK.

How is a run priced?

A run is one top-level agent invocation, however many steps or tool calls it contains. Provider tokens are billed by your provider.

What happens when a verification fails?

The run is marked failed, the action rolls back if it is reversible, and the owning team is paged with the full trace attached.

Is there a free tier?

Yes. 1,000 runs a month, full tracing, one seat, no card required.

Put your agents on the record.

Start free, keep your stack, and see your first trace within the hour.

Start free

Talk to us

Ship agents you can actually watch.

Most teams can build an agent in a week and cannot explain it in a quarter.

Most teams can build an agent in a week and cannot explain it in a quarter.

Most teams can build an agent in a week and cannot explain it in a quarter.

Plan, run, verify. Every time, on the record.

Plan, run, verify. Every time, on the record.

Plan, run, verify. Every time, on the record.

Every request becomes a written plan.

Tool calls execute inside a sandbox you control.

Nothing ships without passing a check.

Everything production asks for on day one.

Everything production asks for on day one.

Everything production asks for on day one.

One screen for every agent you run.

One screen for every agent you run.

One screen for every agent you run.

Keep your stack. Add the record.

Keep your stack. Add the record.

Keep your stack. Add the record.

Built for teams that get audited.

Built for teams that get audited.

Built for teams that get audited.

Teams stopped guessing.

Teams stopped guessing.

Teams stopped guessing.

We cut incident triage from an afternoon to about six minutes. The trace tells you where it went wrong before you finish reading the alert.

We cut incident triage from an afternoon to about six minutes. The trace tells you where it went wrong before you finish reading the alert.

The approval queue is what got us through legal review. Every high-risk action now has a name attached to it.

The approval queue is what got us through legal review. Every high-risk action now has a name attached to it.

Evals run against live traffic now, not a frozen test set. We catch regressions the morning they ship, not a week later.

Evals run against live traffic now, not a frozen test set. We catch regressions the morning they ship, not a week later.

Per-customer spend caps ended our surprise invoices. Finance stopped asking what the agents were doing with the budget.

Per-customer spend caps ended our surprise invoices. Finance stopped asking what the agents were doing with the budget.

Rolling back an agent used to mean a redeploy. Now it is one click, and the run history shows exactly what changed.

Rolling back an agent used to mean a redeploy. Now it is one click, and the run history shows exactly what changed.

Shipping every week, on the record too.

Replay any run on a new model

Replay any run on a new model

Approval queues in Slack

Approval queues in Slack

Spend caps per customer

Spend caps per customer

EU data residency

EU data residency

Plan

Run

Verify

Roll back

On the record

Priced per run, not per seat.

Priced per run, not per seat.

Priced per run, not per seat.

Before you ask.

Before you ask.

Before you ask.

Do I have to rewrite my agents to use Vantage?

Which models and frameworks are supported?

Can we self-host?

How is a run priced?

What happens when a verification fails?

Is there a free tier?

Put your agents on the record.