Skip to content

THE ENGINE BEHIND HAWIYAT

Hawiyat AI Composer

The engine that figures out the best way to do each task.

Give Composer a task. It picks the best AI, pulls in the context from your systems, and hands you back a result it has already checked. No model juggling, no glue code.

EXECUTION LOOP

Every task becomes a run.

Think of Composer as a careful coworker. You hand it a task like “resolve the refund for order 3051 on WhatsApp in Algerian Arabic”, or a report, an invoice, a research brief, a code change. It pulls the order and the policy, picks the right model, and checks the answer before it reaches your customer. Done in seconds, for a fraction of a dinar.

task ▸ "Resolve refund request, order 3051, WhatsApp, Algerian Arabic"RUN 02 · READY
route: automodel: claude-sonnet → gpt-4o (fallback armed)ctx: order 3051 + refund policy · 4ktools: [whatsapp, crm, erp]cost: 0.6 DZDlatency: 340msquality: 0.97policy: 1.0

UNDERSTAND ▸ intent: refund request · sentiment: urgent · language: ar-DZ · order 3051 pulled from ERP

Reply delivered in 11 seconds. Customer kept, no chargeback, ticket closed. Total cost: 0.6 DZD.

Your task can be the next run.

Try it for your business

01

UNDERSTAND

Classifies the task and pulls the business context it needs: CRM records, docs, order history.

02

PLAN

Chooses the model, context window, tools, and workflow for this specific task.

03

ROUTE

Sends the run to the right model and tools. Models are routes, never locks.

04

EXECUTE

Runs across models and systems with automatic fallbacks when a route fails.

05

EVALUATE

Grades every outcome on quality, latency, and cost. Nothing ships ungraded.

06

RESULT

Returns the answer to your system and keeps the learning for the next run.

ANY MODEL · ANY SYSTEM

One layer. Every model. Every system.

Your stack is already wired in: WhatsApp, CRM, ERP, email, databases, n8n, and anything else you already run, each as a route the layer can reach. Models are routes too. Here is the guest list.

THINK · ROUTES IN

GPT· structureClaude· nuanceGemini· speedLlama· open, at costOpen models· self-host

ACT · ROUTES OUT

WhatsApp

role ▸ support, sales, back-office

ACTIVE

CRM

role ▸ records in, answers out

ACTIVE

ERP

role ▸ orders, stock, invoicing

ACTIVE

Email

role ▸ sequences and triage

ACTIVE

Databases

role ▸ grounded, queried, updated

ACTIVE

n8n

role ▸ existing workflows, routed

ACTIVE

Docs & Sheets

role ▸ reports, briefs, shared tables

ACTIVE

API endpoints

role ▸ your other tools, called on demand

ACTIVE

HABITS

What Composer does with every run.

Six habits, one engine. You bring the business problem, Composer handles the how.

Picks the best brain per task

GPT for structure, Claude for nuance, Gemini for speed, open models when you want them at cost. You hand over the task, never the model pick.

route: auto

Grounds every answer in your business

Composer pulls the record, the order, the policy from your systems before it writes a word. Answers come from your data, not a model's memory.

ctx: grounded

Calls tools only when a task needs them

WhatsApp, CRM, ERP, databases, n8n, and anything else you already run. Each run gets the tools it needs and nothing stuck on. The wiring is Composer's job, not yours.

tools: on-demand

Keeps runs alive when a model doesn't

A route slows down or goes dark, Composer cascades to the next best one. No dead runs, no stuck queues, no 3am rebuilds.

fallback: cascades

Grades every single run

Quality, latency, and cost are checked before anything ships. If a result isn't good enough, it doesn't reach your customer.

quality: checked

Learns so the next run costs less

Cache hits, tighter context, smarter routes. Repeated work drops in cost over time, and every dinar shows on the receipt.

cost: trending down

TELEMETRY

What’s live on the layer.

Verified figures from the Hawiyat operations dashboard. Nothing else gets printed.

200+

clients on the platform

100B+

tokens executed through Composer

ONE CONTRACT

The whole stack. One contract.

Composer runs the layer; the rest of the stack runs Composer. Deployment, support, telemetry, and billing: one team, one invoice, in DZD.

Composer

execution engine: routes, runs, and evaluates every task

n8n

workflow runtime: your flows, hosted and supervised

Evolution API

WhatsApp infrastructure: the channel to your customers

Platform

cloud runtime: servers, storage, and databases under one roof

START

Start your first run today.

Pick a Composer plan in DZD, or bring the whole stack. Composer handles the models. You focus on your business.