Skip to main content

Welcome to the Abacus AI Agent Help Doc

Your guide to unlocking the full power of the first god-tier, all-in-one AI agent.

What is Abacus AI Agent?​

Abacus AI Agent is a general-purpose AI agent that doesn't just think — it acts. It builds and deploys full-stack web and mobile apps, tackles complex software engineering on real codebases, and produces presentations, research reports, and data analyses that are ready to share. It connects to your tools — Gmail, Slack, GitHub, Jira, databases, and hundreds of MCP integrations — and runs work autonomously on schedules and triggers, with self-improving tasks that get measurably better over time. When a job needs hands, it operates a real browser or desktop; when a job is big, it fans out a swarm of agents working in parallel. You describe the outcome — the agent plans, executes, and delivers, end to end.

What Can Abacus AI Agent Do?​

Each capability below has its own guide with demo videos:

CapabilityWhat you get
AppsFull-stack websites and web apps — databases, auth, LLM features, one-click deploys, custom domains
Mobile AppsComplete mobile apps from a prompt — preview on your phone and iterate conversationally
Complex CodingMulti-part software systems, team projects, and GitHub workflows — the agent submits PRs for your review
Self-Improving Tasks and AgentsConnect your tools, put agents on schedules and triggers, and let them measurably improve run over run
Presentations (PowerPoint)Polished decks from a topic or your data — then edit them slide by slide with a prompt
Research and AnalysisDeep, cited research from the live web plus code-free analysis and dashboards from your data
Computer Use and AutomationA real browser and desktop, operated by the agent — forms, multi-tab work, and workflows with no API
Voice AgentsAI agents that talk — standalone calling agents, voice in your apps, and voice help desks
Agent SwarmsBig projects run as parallel tasks by a coordinated team of agents

And more, all from the same chat:

  • AI Chatbots: Train assistants on your content, let them take actions through your tools, and share or embed them anywhere.
  • Short Videos & AI Images: Generate platform-ready short videos and stunning images with the industry's top media models.
  • Documents & Writing: Reports, docs, and long-form writing — structured, sourced, and formatted.

Prompting Best Practices​

To get the most out of Abacus AI Agent:

  • Be clear and specific about the task: what you want, how long, and what format.
  • For complex tasks, break down your task into multiple steps first and then start small and build more on top of it for the best results.
  • Mention if you want sources, charts, visuals, or summaries.
  • You can always say "you decide" if you want Abacus AI Agent to make smart decisions.

How Long Does It Take?​

Most tasks take 5–25 minutes depending on complexity. Make sure your system is on and connected to the internet during the task.

Credits & Pro Tier​

  • Each Abacus AI Agent task typically uses 500–1000 credits.
  • On the Basic Tier, you get 3 Abacus AI Agent tasks/month.
  • Upgrade to Pro Tier ($10/month) for:
    • Unlimited task access as long as you have active credits
    • 10,000 bonus credits, totaling 30K credits
    • Unrestricted Abacus AI Agent access with active credits
    • Unrestricted Abacus AI Studio access with active credits, including no conversation cap and no per-conversation credit cap
    • Stronger AI models for richer results
    • Unrestricted access to SuperComputer — your always-on computer for APIs and persistent agents, where you can build, deploy, and host agents, APIs, and databases

For more details, please refer to the Billing FAQs.

Agent Effort Levels​

What are the different effort levels for Abacus AI Agent?​

Abacus AI uses 20+ AI models to give you the best agentic experience that is optimized based on performance, cost and speed. We recommend setting "Auto" as default. However if you want to accomplish hard tasks, try High or xHigh Mode. xHigh mode routes your coding tasks to very expensive but very capable models like GPT 5.5 xHigh and Opus 4.8 xHigh. Use responsibly!

For the very hardest problems, choose Max (Fable 5) — our most powerful effort mode. When Max is selected, every coding and app-building prompt is routed to Anthropic's Claude Fable 5, the strongest coding model available today. Max shines on complex, multi-step engineering work: large refactors, intricate debugging, and building complete applications from scratch. Because Fable 5 is a premium frontier model, Max consumes the most credits of any mode — reserve it for the tasks where quality matters most. Like High and xHigh, your Max selection is remembered across sessions until you change it.

If you want to keep credit usage to a minimum, try xLow Mode. xLow mode uses fast, cost-efficient open-source LLMs — including GLM 5.2 and Kimi 2.7 Code — to handle your tasks at the lowest credit cost. Note that xLow can only be selected when starting a new conversation; to use it, start a new conversation with the xLow mode.

Still Need Help?​

Before reaching out to support, check:

  • Are you using the right category or prompt for your goal?
  • Have you mentioned the output format clearly?
  • Have you connected your tools for automation?

If you're still stuck, contact support — we're here for you.