ChanlChanl
Agent Tools

Give Agents Real Capabilities

Build custom tools that let your AI agents call APIs, run JavaScript, and execute multi-step workflows — all during live conversations.

Start FreeSee Demo

HTTP Tools

Call any REST API during a live conversation. Look up orders, check inventory, update CRM records — your agent acts on real data.

JavaScript Tools

Run custom JS logic for calculations, data transformations, conditional routing, and anything an API alone can't handle.

Multi-Step Tools

Chain multiple actions into a single tool. Validate input, call an API, format the response, and update your system — all in one step.

HTTP Tools

Call any API during a live conversation

Create tools that hit your REST APIs in real time. Agents can look up customer records, check order status, update CRM fields, or trigger any webhook — all while talking to a customer.

  • GET, POST, PUT, DELETE with dynamic parameters
  • Auth header injection (Bearer, API key, custom)
  • Response mapping to agent-friendly output
  • Timeout and retry configuration
Connected Integrations12 active
SalesforceSalesforce
SlackSlack
GoogleGoogle
StripeStripe
HubSpotHubSpot
IntercomIntercom
ZapierZapier
ShopifyShopify
GitHubGitHub
JiraJira
GmailGmail
PostgreSQLPostgreSQL
JavaScript Tools

Custom logic when APIs aren't enough

Write JS functions that run server-side during conversations. Calculate pricing, validate inputs, transform data, or implement business logic that goes beyond simple API calls.

  • Full JavaScript runtime with async/await
  • Access conversation context and memory
  • Return structured data back to the agent
  • Test and debug in the built-in editor
system-prompt.md
v2.3 → v2.4
# System Prompt
You are a helpful customer support agent.
- Be polite and professional
+ Be warm, empathetic and professional
- Response time target: 5 seconds
+ Response time target: 3 seconds
Always verify customer identity.
+ Ask for preferred contact method.
Tests
All passing
CSAT Impact
+2.4%
Deployed
2 min ago
Multi-Step Tools

Chain actions into powerful workflows

Combine HTTP calls, JS logic, and conditional branching into a single tool invocation. Your agent can execute complex workflows without multiple round-trips.

  • Sequential and parallel step execution
  • Conditional branching based on previous results
  • Error handling and fallback steps
  • Share data between steps automatically
mcp-config.json
Live
{
"mcpServers":
{
"chanl":
{
"url": "https://acme.chanl.dev/mcp",
"transport": "sse",
"apiKey": "sk-chanl-...a4f2"
}
}
}
Tools
12 connected
Memory
Active
Knowledge
3 sources
Platform

Explore More
Features

Build

Every channel, one brain

Run one agent or many across voice, chat, and SMS, all working from the same customer record.

Build

Takes real action

Agents call your APIs and databases to finish the customer's request.

Build

Remembers every customer

Persistent memory follows the customer across channels, so nobody repeats themselves.

Build

Knows your business

RAG-powered answers from your docs, FAQs, and knowledge base.

Build

Change it safely

Track prompt changes, compare versions, and roll back safely.

Build

Connects to anything

Connect any tool via Model Context Protocol. Open standard.

Test

Test before customers

Test with 25+ AI personas across edge cases before customers do.

Test

Every call graded

Custom scorecards evaluate accuracy, compliance, and tone.

Test

Realistic test callers

25+ personas with unique emotions, accents, and conversation patterns.

Test

Catch breaks early

Catch regressions before they reach customers. Run tests on every change.

Test

Stays on script

Verify your agent follows policies, scripts, and regulatory requirements.

Observe

Monitor AI live

Live dashboards show every conversation and where each customer stands.

Observe

Watch and step in

Watch conversations live and take over before a customer is lost.

Observe

Full transcripts

Every conversation with speaker labels, timestamps, and sentiment analysis.

Observe

Know when it breaks

Get notified when agents fail or sentiment drops, before customers feel it.

Observe

Know your spend

Track spend per agent, per channel, and per conversation.

Learn agentic AI

Learn how to build agents that work

Watercolor illustration of an observation tower overlooking two parallel worlds, Blade Runner 2049 style in sage and olive tones
Testing & EvaluationApr 15

Is AI Better Than Your Humans? Score Both on One Rubric

Watercolor illustration of two figures walking through a warm corridor of looping paths, Her style in warm plum tones
Testing & EvaluationApr 15

Every Failed Call Is a Test Case You Haven't Written Yet

Watercolor illustration of a space mission control room with signal data flowing across screens, Interstellar style in dusty blue tones
Industry & StrategyApr 15

Your Conversations Are Already CRM Data. Here's How to Use Them.

Open-source AI agent testing engine with conversation simulation and scorecard evaluation
Testing & EvaluationApr 1

We open-sourced our AI agent testing engine

Warm watercolor illustration of a woman at a sunlit flower market, holding a phone to her ear while browsing bouquets
Voice & ConversationMar 21

Customers Don't Trust AI Voices. Here's What Actually Changes That

woman in black long sleeve shirt standing beside woman in gray long sleeve shirt - Photo by Maxime on Unsplash
OperationsMar 20

The AI Agent Dashboard of 2026: What Teams Actually Need to See

Man and woman back to back in office - Photo by Vitaly Gariev on Unsplash
OperationsMar 19

AI Agents Are Great. Until They're Not. When to Put Humans Back in Control

Two men filming a scene outdoors with artwork. - Photo by Luke Thornton on Unsplash
Testing & EvaluationMar 19

Zero-Shot or Zero Chance? How AI Agents Handle Calls They've Never Seen Before

See all articles

Learn Agentic AI

Weekly. Patterns for shipping agents that work. MCP, scorecards, regression tests, prompts, model comparisons.

500+ builders subscribed

Frequently Asked Questions