▦AVAILABLE
ERP
100+ realistic ERP tools across procurement, inventory, AP, AR, sales, payments, approvals, and reporting.
procurementinventorypaymentsapprovals+ more
Start with ERP ↗ YOUR AGENT. A SAFE PLACE TO FAIL.
The simulation
layer for
AI agents.
Test your AI agents inside realistic, isolated business environments. Before they touch production.
$ npm install @burn0/siloCopied to clipboard
BRING YOUR OWN BRAIN.
LangChain
▲ Vercel AI SDK
OpenAI Agents SDK
Anthropic
Mastra
{ custom }
01 / THE ENVIRONMENT LAYER
Your agent thinks.
Silo gives it a world.
Keep your reasoning engine. Silo provides realistic tools, seeded state, tasks, and deterministic verifiers around it.
BRING YOUR OWN
Existing agent
Any framework.
Even a raw API loop.
Silo runtimeISOLATED
SIMULATED BUSINESS WORLD
⌘ToolsRealistic actions
▤StateSeeded data
◎TasksClear objectives
✓VerifiersOutcome checks
tool call → state change → trace → verification
INSPECT THE OUTCOME
✓ PASS reward 1.00
× FAIL with evidence
Traces. State diffs.
Repeatable run artifacts.
Silo does not care how your agent thinks. It gives your agent a world to act in.
02 / ENVIRONMENT TEMPLATES
Real systems.
Zero real-world damage.
Start with a business environment.
Or build a world of your own.
[ + ]AVAILABLE
Blank canvas
Your tools. Your state.
Your definition of success.
world = your_rules;Create an environment ↗CRM
COMING SOONFinance
COMING SOONSupport
COMING SOONProject Management
COMING SOON03 / FROM INSTALL TO EVIDENCE
A new world.
The same agent.
No agent rewrite. No production credentials.
Just connect, run, and inspect.
Initialize environment
Name your environment, choose a template, and add tools.
Connect agent
Pass simulated tools to the agent you already use.
Run rollout
Execute tasks against a repeatable, seeded world.
Verify outcome
Inspect state changes and deterministic evidence.
⌁ terminal
Copied to clipboard
$ npm install @burn0/silo $ silo init ? Environment name procurement-test ? Choose template ERP ✓ Add tools ✓ Connect existing agent $ silo run --env procurement-test \ --task erp-042
TS silo.agent.tsADAPTER
import { createSiloAgent } from "@burn0/silo"; import { runAgent } from "./src/my-agent"; export default createSiloAgent({ run: ({ task, tools, callTool }) => runAgent({ task, tools, callTool }), });
04 / NOTHING HIDES IN THE TRACE
“It worked” isn't enough.
See what actually happened.
Every tool call. Every state change. The exact reason a task passed or failed.
silo / procurement-testHOSTED DASHBOARD PREVIEW
TASK / ERP-042
Create a compliant purchase order
procurement-agent · seed 42 · 4.2s
Reward1.00 / 1.00
Tool calls4
Pass rate · demo87.5%
State changes2
01182ms
vendors.get{ vendor_id: "V-019" } → active: true0296ms
inventory.get{ sku: "SKU-402" } → available: 1203241ms
approvals.request{ amount: 4800 } → approval_id: "APR-88"04310ms
purchase_orders.create{ approval_id: "APR-88", total: 4800 } → PO-1043TEST BEHAVIOR, NOT JUST OUTPUTS.
Does the code work?
Test individual functions.
Does the answer look right?
Evaluate generated outputs.
Did the agent do the right thing?
Test behavior inside a world.
YOUR MACHINE. YOUR WORLD.
Local by default.
Open by design.
↳ Run locally. Keep your agent and data on your machine.
↳ Cloud and hosted dashboards are optional.
↳ Open source. TypeScript-first.
LET IT FAIL HERE. GET IT RIGHT OUT THERE.
Build agents you can trust
before production.
Test behavior, not just outputs.
$ npm install @burn0/siloCopied to clipboard
Get started →