The Agent-Operated AI Console
A desktop console where you say what you want and an agent does the work — turning the apps you already run into agent tools with MCP Forge, testing your product like a senior QA engineer with Agentic Testing, and stretching every token with Spectre-Tree-Grounding so smaller, cheaper models take on heavier work.
Flow Console is a desktop console for getting real work done with AI agents. Its front door is Agent Home: describe the outcome you want in plain language, and a scoped agent operates the console for you — opening the right tools, doing the work in its own tab, and pausing only when you need to sign in, decide, or approve a step that matters. The expert panels are all still there; they just stop being the price of entry.
Under the hood, Spectre-Tree-Grounding keeps agents anchored to a live, structured map of your project instead of re-reading everything each turn — trimming token consumption by up to ~35% in heavy sessions and letting smaller, cheaper models carry work that would usually need a frontier model. Bring your own cloud models, or run local models on your machine when you prefer — the choice is yours.
Point Agentic Testing at your project and a running build, and it tests the way a senior QA engineer would: it learns what the product is supposed to do, drives every button, form, and flow in a real browser, and verifies each expectation criterion by criterion. It works alongside your engineers — drastically reducing the manual testing workload and accelerating every release cycle. The result is a report you can hand to a stakeholder — every verdict cross-checked against what the machine actually observed, every untested control accounted for. Release day stops being a guess.
Describe a capability in plain language, and a Forge agent does the rest: it locates the feature in the knowledge graph and precompiled documentation Flow Console maintains for your application, pins the workflow at the protocol level — real, typed requests, not screen automation — and compiles it into a TypeScript MCP server with clean tool definitions. Verify the result, then run it on your own infrastructure or let Metantel host it for free. Low overhead, agent-assisted throughout, and security is inherited from your application: same login, same permissions, same audit trail.
Flow Console keeps a live, structured map of your project — its files, flows, and how they connect — and grounds every agent step in it, so agents stop burning context re-reading the codebase each turn. In heavy sessions that trims token consumption by up to ~35%, and it lets smaller, cheaper models take on tasks that would normally demand a frontier model. Lower cost per run, faster iterations, and models that punch above their weight — Spectre-Tree-Grounding is part of the broader Spectre-MCK suite we are building for the console. Prefer to keep everything in-house? You can still run local models on your own machine.
At the core of Flow Console is the SPECTRE engine — technology developed within Spectre, Metantel’s advanced-projects division. Its methods are proprietary; its results are the product: it lets the console read an application the way an engineer does, and act on it the way a careful operator would.
~730 MB installer · connect a cloud model in seconds, or download the free local model pack (~11 GB) inside the app if you would rather run on your own machine
Flow Console doesn't ship in isolation. It plugs into the rest of the Metantel line-up through one shared protocol surface — context flows in and out without leaving the family.
Elenchus is the mobile command centre — start a task from your phone, watch it run live, stop it or queue the next one from anywhere.
Open Elenchus →Flow Console agents read project context from Artel and write results back — docs, decisions, and task updates land where your team already works.
Open Artel →