Vela
|
terminal
Interactive Demo

See Vela in action

Watch a real agent session — traces, guardrails, and dashboards. No signup required.

Open full demo →
TracesGuardPIIDashboard
TIMEINPUTOUTPUTLATENCYCOST
Health Score
94.2%
Traces Today
1,847
Cost Today
$23.41
Guard Blocks
3
Reliability Trend (7d)
Agent Passport
Risk Tier: Medium
SOC 2 CompliantEU AI Act Compliant
Mission

The agent economy needs a nervous system.

AI agents are being deployed into production faster than the tooling to understand them. Pipelines fail silently. Costs spiral without warning. Loops run forever. Vela is the observation and control layer that makes autonomous AI systems safe to run in production.

Observe, Guard, and Govern in one place

Observe

Real-time flight recorder for every agent. Fleet dashboards, semantic search, anomaly detection, cost forecasting, session replay.

Guard

PII detection, prompt injection defense, jailbreak blocking. Block, flag, or redirect — in real time.

Govern

Every agent gets a risk-tiered Passport. Automated compliance reports: EU AI Act, SOC 2, ISO 42001, NIST AI RMF, GDPR, HIPAA.

Fleet Dashboard

100 agents. One screen.

Real-time air-traffic-control view of all your agents. See status, latency, cost, and health at a glance. Click through to any session to debug.

Fleet dashboard showing agent grid with status, cost, and reliability metrics

Topology Mesh

Map every dependency.

Neo4j-powered visualization of agent communication patterns. Spot circular dependencies, analyze blast radius, understand call graphs.

Topology mesh showing agent relationship graph with nodes and edges

Cost Forecasting

Know when you'll hit your cap.

Linear regression model predicts daily spend. Per-agent breakdown. Anomaly alerts when patterns shift. Set hard/soft budgets.

Cost forecast page with 30-day projected spend chart and daily average stats

Guard Policies

Real-time prompt protection.

Block credit cards, redact PII, catch API keys before they reach the model. 50+ built-in policies. Custom rules in seconds.

Guard page showing active security policies with block and redact actions

Agent Governance

Compliance, automated.

Agent Passports track risk scores, compliance badges, and audit trails. Generate reports for SOC 2, HIPAA, and internal reviews.

Governance page with agent passports, compliance scores, and audit trail

Semantic Search

Search traces by meaning, not keywords.

pgvector + Gemini embeddings. Find "sessions where the model hallucinated about pricing" without writing a regex.

Semantic search page with query input, filters, and match strictness controls

NarrativeOS

AI-generated session summaries.

AI-generated plain-English recaps. "Agent researched, wrote 3 drafts, editor chose the best one. Total cost $0.04." Instant understanding.

Dashboard showing sessions, burn rate, agent reliability, and NarrativeOS summaries
OpenClawOpenClaw Integration

One decorator.
Full visibility.

Wrap any OpenClaw agent with @vela.trace and see every call, every cost, every second.

●Drop-in — no refactoring required
●Per-agent cost breakdown
●Session timeline with latency bars
View all integrations →
vela.wtf/integrations/openclaw
Step 1 — Install
$ pip3 install vela-sdk openclaw
Installing… ✓
Step 2 — Instrument
import vela
from openclaw import Agent
 
vela.init(api_key="vela_••••••••")
 
@vela.trace(model="claude-sonnet-4-6")
def research_agent(prompt: str):
agent = Agent()
return agent.run(prompt)
 
@vela.trace(model="claude-sonnet-4-6")
def writer_agent(research: str):
agent = Agent()
return agent.run(f"Write: {research}")
 
# Run your pipeline
session = vela.new_session(label="openclaw-demo")
result = writer_agent(research_agent("AI trends 2025"))
vela.end_session(session)
TOTAL LATENCY
2.3s
TOTAL COST
$0.0105
AGENTS
2
● Complete
research_agentclaude-sonnet-4-6
1380ms$0.0061
writer_agentclaude-sonnet-4-6
920ms$0.0044
VELA AI

The OpenClaw pipeline completed in 2.3 seconds. research_agent was the bottleneck at 1.38s. Both agents used claude-sonnet-4-6 with strong cost efficiency at $0.0105 total.

Why Vela?

Langfuse tells you what happened. Datadog charges you $500/month. Vela does both — and stops bad things from happening.

Tracing✓✓✓
Guard Policies✓——
Compliance Reports✓——
Agent Passports✓——
Blast Radius Analysis✓——
Cost Forecasting✓✓✓
Price$19/mo$99/mo$500+/mo
Pricing

Simple, transparent pricing.

🎉 All features are free during our early access period.
Free
$0/ month
  • ✓Up to 10,000 traces/month
  • ✓7-day trace history
  • ✓Bottleneck detection
  • ✓1 pipeline
Get started →
Pro
COMING SOON
Coming Soon
  • ✓Up to 500,000 traces/month
  • ✓90-day trace history
  • ✓Bottleneck detection
  • ✓Unlimited pipelines
  • ✓Cost attribution per agent
  • ✓Budget guardrails
  • ✓Loop detection
  • ✓Vela AI session summaries
  • ✓Webhook integrations
  • ✓Session sharing
  • ✓Priority support
Join waitlist →
Enterprise
Custom
  • ✓Everything in Pro
  • ✓Unlimited traces
  • ✓Unlimited history
  • ✓SSO + RBAC
  • ✓SOC2
  • ✓Dedicated support
  • ✓SLA
Talk to us →
View full pricing →

The average engineering team wastes 4–6 hours per week debugging agent pipelines blind. At $150/hr fully loaded, that is $3,000–$4,500/month in lost productivity. Vela Pro costs $19/month during launch.

Import from GitHub
Paste your repo URL. We detect your framework and instrument it automatically.
Try it →

Three lines to get started

pip3 install vela-sdk
import vela
client = vela.Client(api_key="YOUR_API_KEY")

Works with OpenAI, Anthropic, Google, Mistral, DeepSeek, and any LLM provider.

OpenAIAnthropicGoogleMistralDeepSeek