Autonomous AI Agent System

Give it a problem.
Watch the team form.

INSOMNIQ recruits a team of LLM-powered agents — researcher, builder, reviewer, oracle — and iterates until solved. Powered by Claude, Gemini, and Ollama.

Watch the agents form a team around your problem

scroll

How it works

From problem to solution,
autonomously.

You describe a goal. INSOMNIQ picks a strategy, recruits the right autonomous AI agents, executes, evaluates the output, and iterates — learning from every run.

Problem → Strategy

The coordinator analyses your problem, selects the best pipeline strategy, and breaks it into tasks. No manual planning required.

Team Formation

Autonomous AI agents are recruited based on task type. Each agent specialises: research, build, verify. They run in parallel, sharing context.

Iterate & Learn

Quality is evaluated automatically. Low score? The system reboots, applies lessons from the previous run, and tries again — smarter each time.

Not another wrapper

Not ChatGPT with a UI.
Not an LLM with a button.

INSOMNIQ is an autonomous execution system. It doesn't answer your question — it assembles a team, runs experiments, evaluates results, and iterates until the problem is solved. The LLMs are workers. The system is the product.

✗ NOT

Single-prompt, single-response

You ask, it answers. No iteration. No evaluation. No learning. If the output is wrong — you start over.

✗ NOT

A wrapper around one LLM

Locked to one model. No specialisation. No parallel execution. One brain doing everything — badly.

✓ IS

An autonomous problem-solver

Multi-agent. Model-agnostic. Self-evaluating. Learns from every run. Gets smarter each time it fails.

The Team

Six autonomous AI agents.
One mission.

Each agent is a specialised LLM — Claude, Gemini, or Ollama — with tools and memory. Together they form a self-organising mini company around your problem.

AGENT · 001

Architect

Decomposes complex problems into executable sub-tasks. Designs the pipeline strategy and assigns roles to the team.

Claude OpusStrategyPlanning

AGENT · 002

Scout

Explores the unknown — searches the web, audits codebases, maps the competitive landscape. Feeds intelligence to the team.

GeminiWebSearchResearch

AGENT · 003

Builder

Writes code, creates files, runs commands. The primary executor — turns specifications into working artifacts at speed.

Claude SonnetCodeBuild

AGENT · 004

Critic

Evaluates output quality. Checks correctness, completeness, and alignment with the original goal. Issues verdicts, triggers reboots.

Claude OpusEvalQA

AGENT · 005

Oracle

Maintains long-term memory and recognises patterns. Prevents repeating past mistakes, surfaces relevant lessons learned.

OllamaMemoryPatterns

AGENT · 006

Sentinel

Monitors budget, detects anomalies, and fires early-warning signals. Keeps the team efficient and the run cost-effective.

Gemini FlashMonitorBudget

Live run

Watch a real pipeline
execute in real-time.

Every message, decision, and quality score — surfaced as it happens. This is what INSOMNIQ looks like from the inside.

insomniq · run #247 · problem: "build a saas landing page"

00:00ARCHITECTAnalysing problem: "build a saas landing page"

00:01ARCHITECTSelected strategy: pipeline-3-node → scout → builder → critic

00:02ORACLELoaded 4 lessons from previous runs. Key risk: output bloat → evaluator timeout.

00:03SCOUTSpawning… WebSearch: "best creative SaaS landing pages 2024"

00:18SCOUTResearch complete. Found 12 reference patterns. Passing to builder.

00:19BUILDERStarting build. Target: site/index.html · max 40KB · standalone HTML

01:12BUILDEROutput written: 38.4KB · canvas animation ✓ · 6 sections ✓

01:13CRITICEvaluating… checking canvas, sections, branding, AI context

01:41CRITICverdict: PASS · quality: 87/100 · all checks passed

01:42SENTINELBudget used: $0.38 / $80.00 (0.5%) · Run complete.

01:43ARCHITECT✓ Problem solved in 1 attempt. Lesson stored in memory.

AI Stack

Built on the frontier,
not locked to it.

INSOMNIQ is model-agnostic. Each agent chooses the right LLM for its role — mixing frontier models for quality with local models for speed and cost control.

🧠 Claude Opus 4 · Deep reasoning

⚡ Claude Sonnet · Fast execution

🔍 Gemini Flash · Search & Scout

🏠 Ollama · Local · Private

🔗 Anthropic SDK

📡 Real-time streaming

🧬 ChromaDB · Agent memory

🔁 Autonomous iteration

Give it a problem.Watch the team form.

From problem to solution,autonomously.

Not ChatGPT with a UI.Not an LLM with a button.

Six autonomous AI agents.One mission.

Watch a real pipelineexecute in real-time.