A private autonomous system that executes code, controls browsers, sends messages, manages files, and remembers everything — operating at a level most platforms won't even attempt.
> autonomous execution / persistent memory / browser control / desktop automation / zero restrictions
Invoker doesn't suggest. It executes. Browser sessions, file operations, code deployment, message delivery, desktop control — full stack autonomous operation.
Multi-iteration task execution powered by Claude Code. Submits work, supervises output, auto-repairs failures, verifies results, and delivers structured proof — fully autonomous loop.
Persistent context engine across every session, project, and conversation. Manual and automatic memory. Project-level recall. Nothing is ever forgotten unless you want it gone.
Headless Playwright with your real Chrome session — all cookies, all logins. Reads dashboards, fills forms, extracts authenticated data, clicks through interfaces. Full web agency.
WhatsApp message delivery with generated images, contact management, and conversation handling. Tell it who to message and what to send — text, images, or both.
Five AI models under one roof. Qwen 32B local for privacy and speed. GPT-5 and Claude for deep reasoning. DALL-E for image generation. Automatic routing — the system picks the weapon.
Direct screen control — mouse movement, keyboard input, application switching, UI interaction. When browser automation isn't enough, the entire desktop becomes the interface.
DALL-E 3 integrated at prompt level. Generate images inline, attach to messages, embed in workflows. One prompt creates and delivers.
Every operator task runs through a supervisor layer — OpenAI reviews Claude's output, decides continue/done/blocked, and triggers repairs when verification fails. Trust but verify, automated.
Upload documents, build project memory, carry structured context across sessions. File summaries, mechanics extraction, and auto-retrieval — the system knows what you're working on.
Every task enters a disciplined pipeline. No guessing. No hoping. Classify, execute with the right engine, verify the output, repair if needed, deliver with proof.
Neural classifier analyzes intent — chat, operator task, browser job, image generation, messaging, desktop control — and dispatches to the optimal engine in milliseconds.
Claude Code operates with full tool access. Edits code, calls APIs, browses authenticated pages, generates images, sends messages, controls the desktop. Real tool use, not text completion.
Independent supervisor reviews every output. Failed verification triggers auto-repair. Regression tests run. Structured summary delivered with cost, duration, files touched, and terminal verdict.
Not a wrapper. Not a proxy. A complete private system with its own backend, worker pool, database, browser engine, model routing, and memory architecture.
Your backend. Your workers. Your data. Your rules. Local models handle fast private work. Cloud models escalate when depth is needed. The entire pipeline runs on your machine — no third-party sees your prompts, no platform decides what you're allowed to do.
Qwen 32B running local. GPT-5.2 for peak reasoning. Claude Sonnet for operator execution. Claude Opus for supervision. DALL-E 3 for generation. Automatic selection based on task weight.
Multiple workers polling in parallel, claiming tasks instantly. Job queuing, heartbeat monitoring, stale recovery, crash resilience. Your operator never sleeps.
Every operator session logged with events, iterations, supervisor decisions, cost tracking, file changes, and terminal verdicts. Complete transparency. Zero black boxes.
Custom instructions, personality presets, tone profiles, restriction levels, voice read-aloud. Dial the system to match exactly how you think and work.
The system is live. The workers are running. The models are loaded. Enter and operate.
Enter Invoker 2