Now supporting Claude 3.5 & GPT-4o

See every dollar your AI agents spend

One URL change. Monitor costs in real time, set hard limits that stop runaway spend, and get AI-powered recommendations to cut your bill by 40%.

app.nolma.ai/dashboard/guard
N
Nolma
GUARD
Overview
Sessions
Budgets
Alerts
LENS
Overview
Recommendations
Shadow Mode
SHIELDsoon
L
lumyteai

Guard Overview

Today · syncing...
⏸ Pause agents
Live

$2.84

Total cost today

↑ 12% vs yesterday

1.2M

Tokens consumed

↓ 3% vs yesterday

847

Sessions today

3 agents running

73%

Daily budget used

Spend over time
1H6H24H
⚠ spike2am4am6am8am10am12pm4pmnow$0$1$2$3
Burn rate: $0.12/hr
Projected month: $86.40Under budget
Top agents
support-bot$187.23
4,821 callsgpt-4o
code-reviewer$108.40
2,103 callsclaude-sonnet
report-generator$52.10
847 callsclaude-haiku
Budget enforcement
support-bot · $15/day73%
code-reviewer · $10/day45%
org-wide · $50/day21%
💡 Lens

support-bot acceptance

89.3%

downgrade readiness

78/100

est. monthly saving

$43.20

Switch to gpt-4o-mini→ Save $43/mo

Works with every LLM provider and AI framework

O
OpenAI
A
Anthropic
G
Gemini
G
Groq
M
Mistral
L
LangChain
Ll
LlamaIndex
C
CrewAI
O
OpenAI
A
Anthropic
G
Gemini
G
Groq
M
Mistral
L
LangChain
Ll
LlamaIndex
C
CrewAI
A
AutoGen
Vercel AI SDK
OR
OpenRouter
T
Together AI
F
Fireworks AI
P
Perplexity
C
Cohere
R
Replicate
A
AutoGen
Vercel AI SDK
OR
OpenRouter
T
Together AI
F
Fireworks AI
P
Perplexity
C
Cohere
R
Replicate

The platform

Everything your AI stack needs.

Three products that work independently or together. Start with Guard. Add Lens when you want quality insights. Add Shield when compliance matters.

Guard

Real-time cost control

Hard budget enforcement fires before the LLM call — zero cost incurred on blocked requests. Track spend per agent, model, environment, user, and session.

  • HTTP 429 before any API call fires
  • Auto model downgrade on soft limits
  • Anomaly detection: spikes & retry loops
  • Weekly AI cost digest via Claude Haiku
Lens

Prompt quality intelligence

Every user interaction is a signal. Measure acceptance rates, edit distances, and retry patterns to understand which prompts and models actually work.

  • Downstream user signal collection
  • Evidence-backed prompt rewrite suggestions
  • Downgrade readiness score (0–100)
  • Shadow mode: safe 10% traffic splits
Shield

Security & compliance

Block prompt injection in <0.1ms. Redact PII before it leaves your stack. Produce immutable GDPR / OWASP / EU AI Act audit trails on demand.

  • Regex + DeBERTa semantic injection detection
  • PII redaction via Microsoft Presidio
  • Output content moderation (0ms impact)
  • One-click compliance PDF exports

The problem

Agents don't care about your budget.

Without observability, every LLM agent is a cost black hole. One retry loop and you're explaining a four-figure bill with no breakdown to show for it.

Without Nolma
your-agent.log
[ERROR] Unexpected OpenAI charge: $1,247.83
[ERROR] Expected monthly budget: $50.00
 
[WARN] No agent breakdown available
[WARN] No model usage data found
[ERROR] Rate limit hit. Retrying... (47/50)
[ERROR] Rate limit hit. Retrying... (48/50)
 
# Which agent? Which model? Which user?
# You have no idea.

“Saved us $2,400 in the first week. We had a retry loop we didn't even know about.”

— Beta user, early access
Gateway architecture

One gateway.
Every provider.

Your agents don't change. Your providers don't change. Nolma sits in the middle — invisible, instant, essential.

Connects your agents to:

OpenAI · Anthropic · Gemini · Groq · Mistral

8.4M+

Tokens tracked

< 5ms

Gateway overhead

5

LLM providers

$0

Cost when blocked

5ms
Added latency
P99 — Redis check + proxy hop
99.9%
Uptime SLA
Railway persistent server
5
LLM providers
OpenAI · Anthropic · Gemini · Groq · Mistral
$49
Hard enforcement
Per month on Pro plan

Works with every major LLM provider

Replace your SDK endpoint. Keep your existing code. Nolma acts as a lightweight proxy that monitors and optimizes every token without changing how you build.

OpenAIAnthropicGoogle GeminiGroqMistral
main.py
import openai
 
client = openai.OpenAI(
    api_key="sk-...",
    # Just one line to change
    base_url="https://gateway.nolma.ai/v1"
)
 
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}]
)
Proxy Active

Pricing

Transparent pricing

Start free. Upgrade when you need more agents or advanced features. No per-seat fees, no hidden costs.

Starter
$0forever

For indie devs and small projects getting started with AI cost visibility.

Start free
  • 100K tokens / month
  • 3 agents
  • Nolma Guard — cost tracking
  • 7-day log retention
  • Email alerts
Most popular
Pro
$49/ month

For growing teams that need full cost control and prompt quality intelligence.

Start Pro trial
  • Unlimited tokens
  • 10 agents, 3 seats
  • Guard + Lens
  • Slack alerts
  • Hard enforcement
  • Shadow mode
Team
$149/ month

For teams that need security, compliance, and per-user isolation.

Start Team trial
  • Unlimited agents
  • 10 seats
  • Guard + Lens + Shield
  • PagerDuty + OTel export
  • PII redaction
  • Per-user isolation
Business
$399/ month

For regulated organisations that need compliance exports and audit trails.

Start Business trial
  • 25 seats
  • Full Shield + PII
  • GDPR / EU AI Act exports
  • Compliance PDF reports
  • 90-day log retention
Enterprise
Custom

For security teams that need credential monitoring, compliance exports, and advanced threat detection.

Talk to sales →
  • Unlimited seats
  • Shield Advanced
  • Compliance export (GDPR / EU AI Act)
  • Credential sharing detection
  • Privilege drift reports
  • SSO / SAML
  • Custom PII entities
  • Dedicated SLA + onboarding

All plans include a 14-day trial of Pro features. No credit card required.

FAQ

Common questions

Does this add latency to my LLM requests?+

Practically none. The Redis budget check is a single cached GET (<1ms). For Shield, Layer 1 regex runs in <0.1ms. Streaming responses are forwarded chunk-by-chunk with no buffering — time-to-first-token is unaffected.

How does budget enforcement work?+
Which LLM providers are supported?+
Is my prompt text stored?+
What happens if Nolma goes down?+

Get started

Never get a surprise LLM bill again.

Start free in under 5 minutes. One URL change, full observability.

Free plan forever · 14-day Pro trial · No credit card required