Your agents send emails, create tickets, charge credit cards, and call APIs. AgentLedger logs every action, tracks every cost, and kills agents when things go wrong.
import AgentLedger from 'agentledger';
const ledger = new AgentLedger({
apiKey: process.env.AGENTLEDGER_KEY
});
// Group related actions into a trace
const traceId = AgentLedger.traceId();
const { result } = await ledger.track({
agent: 'support-bot',
service: 'slack',
action: 'send_message',
traceId,
input: { channel: '#support' },
captureOutput: true,
}, async () => {
return await slack.chat.postMessage({
channel: '#support',
text: 'Issue resolved!'
});
});Works with LangChain · OpenAI Agents · CrewAI · MCP Servers · Express · Python & TypeScript
Live Preview
This is what your dashboard looks like when agents are running. Every action logged, every cost tracked.
Features
LLM observability tools track token usage. AgentLedger tracks what agents actually do.
Real-time SSE streaming of every action. Watch your agents work live with sub-second latency.
Define rules or apply pre-built templates: conservative, cost-conscious, compliance. Rate limits, allowlists, cost caps, and more.
Require human approval for high-risk actions. Approve or deny from the dashboard, Slack, Discord, or PagerDuty.
Multi-day trend analysis, cost forecasting with linear regression, and budget overrun predictions.
Step through agent traces action-by-action. Inspect input/output at each step with a visual replay timeline.
Python & TypeScript SDKs, CLI tool, plus LangChain, OpenAI, MCP, CrewAI, AutoGen, LlamaIndex, and Vercel AI SDK.
Get Started
No infrastructure. No config files. Just wrap and go.
npm install agentledgerawait ledger.track({
agent: 'my-bot',
service: 'gmail',
action: 'send_email',
traceId,
captureOutput: true,
}, sendEmailFn)→ agentledger.co/dashboardComparison
Helicone and Langfuse track LLM calls. We track what happens after.
| Feature | AgentLedger | LLM Tracers |
|---|---|---|
| LLM call tracing | — | ✅ |
| Real-world action logging | ✅ | — |
| Policy engine & guardrails | ✅ | — |
| Human-in-the-loop approvals | ✅ | — |
| Statistical anomaly detection | ✅ | — |
| Agent evaluations & scoring | ✅ | — |
| Trace replay & debugging | ✅ | — |
| Cost forecasting & analytics | ✅ | — |
| Budget controls & enforcement | ✅ | — |
| Batch logging & data export | ✅ | — |
| CLI tool (npx agentledger) | ✅ | — |
| Slack, Discord & PagerDuty alerts | ✅ | — |
| Live SSE streaming | ✅ | — |
| Python & TypeScript SDKs | ✅ | ✅ |
| 8 framework integrations | ✅ | ✅ |
Fully auditable codebase on GitHub
Run on your own infrastructure
Never blocks your production agents
SDK adds no bloat to your project
Pricing
Start free, scale when you need to.
Free tier includes 5,000 actions/month with 7-day data retention. Usage beyond plan limits is rate-limited. We reserve the right to enforce fair use policies to maintain service quality for all users.
FAQ
Know exactly what your agents are doing, what they cost, and how to stop them when things go sideways.
Get Started Free →