Skip to main content

Product Highlights

This page summarizes what ConsoleX does differently. Most of it is about tasks: work you got right once and now run again with checks. The chat workspace you may already know is still there, too.

Every run is checked, with evidence​

Each task has acceptance criteria: what counts as done. When a run finishes, ConsoleX checks the result against them automatically, including the first trial run.

  • Rule checks compare the result text against a rule, such as required sections. They make no model call and cost nothing extra.
  • AI checks have a model review the remaining items after each run, in a single call billed to your plan.
  • A task can have up to 10 acceptance metrics. You can generate them from the task, pick presets from the metric library, or write your own.

Each check gets a verdict, a short explanation and View evidence, with failed and needs-review items listed first. As the report itself says: "Checks use the available output and tool records. A check without enough evidence is marked as needs review, never passed."

Check result with a 4 passed, 1 failed summary and the failed check listed first

Learn more: Read check results and Acceptance criteria.

You make the call on each result​

A check report never decides for you. Under Your call you choose Accept, Needs edit, Reject or Check is wrong, and can add a note. Your decision is saved with the report, and your feedback and the automated verdict are kept separately. Runs that are waiting for your decision appear under Needs you on the Dashboard.

Suggested changes you approve, with version history​

When a run misses, you don't rewrite the prompt by hand. Failed checks, your feedback and your own requests become suggested changes to the task:

  1. Describe what to change, and optionally reference a problem run so its check findings are included.
  2. Review the suggestion as a before-and-after comparison, and accept or reject each change.
  3. Apply it. Later runs use the new version and are checked against the same standard.

A suggested change shown as Before and After, with Accept and Reject for the change

Nothing changes until you approve it. Earlier configurations are kept as configuration records, and each save of the acceptance criteria is a new version, so you can compare versions and restore an earlier one.

On Pro and Max, you can also turn on Suggest fixes automatically: when an automatic check fails or needs review, ConsoleX drafts a fix for you to review. On every plan you can ask for a suggestion at any time. See Improve tasks.

Estimated cost on every run​

Every run records the model it used, its usage and its estimated cost. The run list shows the estimated credits for each run, and each run's statistics break down tool calls, total tokens, known model and tool fees, duration and time to first token. What the checks cost is listed separately.

Run history with each run's check result, estimated credits and date

A run that has already started finishes, and failed tool calls are not charged. See Credits overview.

Templates with a built-in definition of done​

Task templates give you a working starting point with checks already written. Templates for one-off work include Edit to your rules, Deliverable from your materials, Content repurposing, Sourced comparison, Topic research, Personalized outreach, Document summary and Meeting notes & actions. Templates for regular work include Client report, Weekly competitor brief and Hacker News digest. You can also create a custom task from scratch.

Task template gallery with templates such as Edit to your rules and Deliverable from your materials

Before you save, the template flags what is missing in place: blanks to fill in and any connection or key it needs. Save and run a trial shows you a checked result right away, and a scheduled task only starts its schedule after you have seen the result. See Create a task from a template.

Run it again with new materials, or on a schedule​

From a task's Run menu you can Run now, or Add new materials and run: the saved instructions stay the same, and the new files or text are used only for that run. You can also put a task on a schedule (hourly, daily, weekly, monthly or yearly) and get an email when it finishes or fails. Every run is checked the same way.

The Free plan includes 1 scheduled task; Pro and Max include unlimited scheduled tasks. See Run tasks.

Use your tasks in Claude Code​

On Pro and Max, you can have Claude Code do a ConsoleX task on your computer. ConsoleX checks each delivery against the task's saved acceptance standard and records every attempt. If a check fails or needs review, the agent reports it and waits for your decision before making changes.

Run this task in Claude Code dialog explaining how each delivery is checked

See Coding agents.

A full AI workspace​

The chat workspace is where you work things out before saving them as tasks:

  • Any leading model. Auto picks a suitable model for you, or you choose one from the model picker. Pro models are available on Pro and Max. You can also bring your own API key; those model calls are billed by your provider instead of your plan.
  • Compare generation. Regenerate any reply with another model and see the results side by side, or Fork from here to branch the conversation.
  • Tools and MCP servers. About 120 tools for searching, reading files and pages, running code, creating media and reaching services such as Notion or GitHub. Add your own MCP server or HTTP API tool. See Capabilities overview.
  • Skills. Preset Skills for ways of working such as document handling and building artifacts, plus Skills you create or upload. See Use preset Skills.
  • Agents. Combine a model, a prompt, Skills and tools into an agent and reuse it in chats and tasks. Eight preset agents are ready to use. See Use preset agents.
  • Artifacts. Web pages, React components, SVGs, diagrams and documents built in a chat are saved privately in your Library. See Artifacts preview.
  • Cloud browser. Open pages that need scripts or clicks in a real browser. See Cloud browser.