# Build AI Agents That Actually Work In Production

Everything you need to build, ship, and improve production-grade AI agents - in one operating system.

## From idea to production – on one Agentic AI OS

Build, ground, test, deploy, and improve AI agents without stitching together a fragmented stack.

1. ### Define Your Agent
   LLM, instructions, knowledge, tools, guardrails - set your agent's identity on the canvas.

2. ### Expand Capabilities
   Flip the toggles that matter - memory, citations, voice, reflection, file uploads, browser.

3. ### Test & Refine
   Run test datasets, inspect evaluation scores, preview every version before promoting it.

4. ### Deploy & Monitor
   Publish to API, webhook, embed, or A2A. Watch run history in real time. Export data.

5. ### Iterate & Scale
   Continuous evaluation catches regressions. Version control owns every release. Sub-agents scale infinitely.

## Your agent begins here.  With tools, memory & knowledge

Use the agent builder to configure behavior, connect knowledge, add tools, assign sub-agents, and apply guardrails – all from one control surface.

🤖 Orchestration  
Configure how the agent plans, routes, delegates, and coordinates work across multi-step and multi-agent flows.

📋 Knowledge  
Ground the agent in the business context it needs to retrieve, reason, and respond with relevance.

🗃️ Tools  
Connect APIs, systems, and workflows so the agent can take action, not just generate responses.

## Turn on the capabilities your use case actually needs.

Every agent starts the same. What makes yours extraordinary is what you turn on. Eleven capabilities that take your agent from functional to formidable.

👋 ### Welcome Message  
Set the tone from the first word. A custom greeting that makes users feel like they're talking to your brand, not a bot.

🧠 ### Context Window  
Control how far back your agent looks. Deeper memory or leaner cost - you decide the tradeoff.

📎 ### Citations  
Every answer comes with a source. Builds trust, reduces hallucinations, and makes your agent auditable.

🔗 ### Tool Citations  
When your agent pulls data from a tool, it says so. Full transparency, end to end.

📂 ### File Upload  
Users share a contract, a spreadsheet, a screenshot. Your agent reads it in context and responds with precision.

💡 ### Follow-up Suggestions  
After every response, your agent asks the right next question. Conversations go deeper with less effort from the user.

🌐 ### Browser Use  
Your agent doesn't just know - it looks things up. Live web browsing and page interaction, built in.

📋 ### Artifacts  
Generate tables, code blocks, and structured documents inline. Outputs your users can actually use.

🎤 ### Voice  
A full voice interface - transcription, synthesis, and interruption. Your agent speaks in the voice you choose.

🔄 ### Reflection  
Before it answers, it checks its own work. Self-correction before delivery means fewer embarrassing moments.

💾 ### Memory  
Your agent remembers who it's talking to. Preferences, history, context - not just the last message.

🛡️ ### Guardrails  
Block competitors, enforce tone, apply compliance rules. Your brand stays protected in every reply.

## An Agent that remembers.  With memory & knowledge graphs

Reliable agents need more than prompts and documents. SimplAI gives agents persistent memory and a knowledge graph so they can retain context, understand relationships, and respond with greater precision over time.

⚡ Persistent Memory  
Retain conversation history, preferences, and key context across sessions so the agent responds with continuity, not repetition.

🏛️ Connected Knowledge  
Model entities and relationships across your business so agents can retrieve information from structured context, not isolated chunks alone.

🎯 More Reliable Responses  
Combine memory and connected knowledge to reduce hallucinations, improve relevance, and make agents behave more consistently in real workflows.

## Voice AI agents for real conversations, not transcripts.

Configure transcription, synthesis, interruptions, and tool-aware responses so your agent can handle live voice interactions with speed, context, and control.

🎤 Real-time transcription  
Capture spoken input as it happens so the agent can respond without losing context.

🔊 Natural voice synthesis  
Control how the agent speaks back with clear, consistent, brand-aligned voice output.

⏸️ Interruption Handling  
Let users pause, redirect, or clarify naturally without breaking the flow.

🔧 Tool-aware responses  
Allow the agent to fetch information, trigger actions, and complete workflows during the conversation.

## Deploy AI agents anywhere  Zero compromise.

Expose agents through APIs, webhooks, embedded widgets, MCP, and future interfaces – without changing how they run.

🔌 REST API  
Direct integration, full control  
`POST/chat/sendSend message`
`GET/chat/replyPoll reply`
`GET/conversation/{id}Full thread`

```bash
# Auto-generated · copy and run
curl -X POST "https://api.simplai.ai/chat" \
  -H "Authorization: Bearer $API_KEY" \
  -d '{"message":"Hello","agent_id":"AGT_x"}'
```

🔗 Webhooks  
Push-based, no polling  
SimplAI pushes results to your HTTPS endpoint the moment a run completes. No polling loops. No missed events.

```json
// Payload hits your URL instantly
{"event": "run_complete", "agent_id": "AGT_abc", "run_id": "RUN_xyz", "output": "Your answer...", "latency_ms": 1240 }
```

🔒 Embed Widget  
One script tag. Any website.

```html
<!-- Paste anywhere -->
<script src="https://app.simplai.ai/embed.js" data-agent="AGT_xxx" data-token="TKN_xxx"></script>
```

🤝 MCP & A2A  
Protocol v0.3.0 · JSONRPC  
Enable seamless communication between agents using MCP and A2A protocols.

## Every Conversation. Every Run. Immutable Traces.

Trace every interaction end to end including latency, cost, tool calls, and decision flow – so you can debug faster, optimize performance, and understand exactly how your agent behaved in production.

⚡ Performance Metrics  
Avg latency, cost per run, tool call time, steps taken, context utilization, prompt and completion token counts - all at a glance.

🔍 Deep run inspection  
Click any run to see the full conversation trace - inputs, outputs, tool calls, timestamps, and which version ran it.

📊 Filter and export  
Filter by status, user, or date range. Download your entire run history for offline analysis, compliance audits, or reporting.

## AI agent testing & evaluation, from pre-launch to production

Run structured evaluations before launch and continuously score production traces to catch regressions before users do.

✏️ Test Datasets - pre-launch  
Run your agent against a curated dataset. Use AI to generate test cases, set evaluator criteria (accuracy, structure, completeness), and measure before you ship.

📈 Evaluate - continuous quality  
Point the evaluator at past or future conversations. Get average, min, and max scores across thousands of traces - automatically.

🔬 Custom evaluators  
Build scoring criteria specific to your use case - tone, policy compliance, factual accuracy. Your domain, your rubric.

## Every release, under your control.

Label, preview, promote, and roll back any agent release with confidence – without losing visibility over what changed, when, and why.

📤 Publish with confidence  
Give every release a name and changelog. Your team knows exactly what changed and when.

⏪ Instant rollback  
Promote any previous version to production in one click. Production incident? Fixed in under 30 seconds.

👁 Preview before you publish  
Chat live with any version before setting it as latest. Ship what you've tested - not what you hope works.

## Your next agent starts right now.

The canvas is waiting. The features are ready. All that's missing is the idea you've been sitting on.