Master Gridable.ai

Gridable.ai is an AI-native operating system for work. It gives you one console — workflows, connections, assets, monitoring and the apps you build — and AI agents that help you operate all of it. Unlike traditional tools, Gridable.ai is built from the ground up to be orchestrated by intelligence, not just operated by humans.

(1) Sign up with your email. (2) Your workspace is created for you. (3) Ask the Planner, in the console's chat, for the workflow or app you need. (4) Review what it builds, run it, and find every run on the record.

Everything in Gridable.ai revolves around the Workspace and its Views. A Workspace holds your work; Views are the surfaces that operate it — Workflows, Connections, Assets, Monitoring, Schedules and the Apps you build. Workflows are authored in a visual editor where you drag and connect nodes. Agents are the autonomous or semi-autonomous intelligences that live in your workspace and can act across those views on your behalf.

Type an at sign and a name to address a coworker or a person: the message goes to that coworker, or is filed for that person. Type an at sign and a team's name and the message goes to that team's lead, who may hand it on within the team. Type a slash for a command — build, assign, routine, app, find, voice, report, add-to-project, open, help. Type a hash and a project's name in the console's chat and your coworker reads that project's open cards before it answers; a space you are not in stays private, and a name it cannot place gets one question back rather than a guess. In the console's composer, an at sign or a hash opens a menu that narrows as you type: arrows move, Enter or Tab picks, Escape closes it, and a space leaves your words as typed. Picking a person, a coworker, a space or a project writes the plain name into your message, so it reads the same from Telegram or by voice. Picking a report, a file, an automation or an app writes a reference to that item instead of its title. In the conversation each reference shows as a chip you can click; one that names something you may not read reads not available to you, and shows no title. A workflow's Post to a space step takes a hash and a space's name. Every card has a key: its project's letters and a number, such as SHP-12. The key stays the same when the card is renamed or moved, so you can type it or say it anywhere, from Telegram or by voice too, and your coworker reads that card. In the console a card's label reads #project › key and title, and picking a card from the menu writes its key.

Type	It names	Example
@	a coworker or a person — the message goes to that coworker, or is filed for that person	@ops-bot, @alice
@	a team — the message goes to the team's lead, who may hand it on within the team	@growth-team
/	a command: build, assign, routine, app, find, voice, report, add-to-project, open, help	/assign
#	a project, in a message — your coworker reads its open cards first	#shopify
#	a space, in a workflow's Post to a space step	#billing
@	the composer menu: people, coworkers and teams, narrowed as you type	@al
#	the composer menu: spaces you are in, projects, open cards, reports, files, automations, apps and connections, each in its own group	#sho
#	a report, file, automation or app picked from the menu — it shows as a chip with its name for anyone who may read it	#Q3
#	a card, by its key — letters and a number that never change; the hash is optional	SHP-12, #SHP-12

Gridable.ai runs as one service with a per-workspace event bus. When a run is created, a task starts or finishes, or a step is recorded, that event is first written to the run's durable timeline and then pushed to whichever browsers are open on that workspace over a WebSocket. Persist-then-broadcast is the point rather than an optimization: a console that was shut when the event fired still shows it on reload, because the stored timeline is the source of truth and the live frame is only a shortcut to it. Agent handoffs are ordinary steps on that same timeline, so a multi-agent run is auditable afterwards and not merely watchable while it happens.

Each agent has a role, the tools it may use and the model it runs on — or smart routing picks the model for each step. An agent cannot send, publish or pay on its own: those actions wait for approval from the person who asked, or a builder.

An agent's description and role shape how it works. Describe the job in plain words when you ask for it; the Planner asks you about any decision it needs before it builds.

Agents keep the conversation they are in, and can save notes to your workspace's long-term memory and recall them in later work.

Agents use the tools of the connections you add: apps connected over OAuth through the connection broker, and remote MCP (Model Context Protocol) servers you connect.

The Gridable.ai Studio includes a visual nodes-and-wires editor. You can design pipelines where data flows between agents and external systems. Triggers can be manual, scheduled, or webhook-based.

Start a workflow by hand, on a schedule, or from a webhook — anything that can POST can start a run. A workflow can hand a step to an agent and branch on a condition.

Clean, filter, and transform data as it moves through your workspace. Use the transform node to reshape data between steps, or a code node for the cases it does not cover.

By default, every request Gridable makes for your workspace through OpenRouter carries OpenRouter's data-collection 'deny' setting: only a host that does not store or train on prompts may answer it, whether the request uses Gridable's key or your workspace's own. If no host of a model qualifies, Gridable moves a smart-routed request to another model in the same tier and says so in the log, and a request pinned to that model fails and says why; it never falls back to a host that trains. Claude and OpenAI models called directly run under Anthropic's and OpenAI's API terms, which do not use API data for training. Questions Gridable's decision model answers about your work go only to hosts that keep nothing, whatever your setting. A workspace admin can turn on "Allow AI providers to train on this workspace's data" in Settings » AI accounts. It asks for confirmation first, the workspace then accepts that choice, and the change is recorded with who made it and when. Members see the setting but cannot change it.

Every step can retry up to five times with a delay between attempts, stop at a timeout, and either halt the run or skip and carry on. A failure is recorded on the run's timeline, and a workflow that keeps failing is escalated to your workspace's admins.