Join our agent building community Discord →

From Pointer to Agentbase

By Thomas Schlossmacher

Direction

A Serverless Platform for Agents

Over the past year we built Pointer thrice: first as a fast, minimal command‑line agent for developers, then as a browser agent to navigate the web, and finally as a focused GUI agent to make the best parts of that experience accessible to teams. Those three iterations taught us the same lesson from different angles: agents need a reliable place to live — to run, scale, observe, and integrate with the rest of a product — without duct tape.

Early 2025: The CLI Agent

Pointer began on the command line. It was opinionated, scriptable, and great for developers. We optimized for speed, deterministic runs, and a clean interface to tools and data. But as soon as agents did anything useful, the real work started: scheduling background runs, handling webhooks, storing state, securing secrets, logging traces, retrying failures, and sharing results with teammates. The moment an agent left a laptop, we were rebuilding infrastructure.

Mid 2025: A GUI for teams

We then built a GUI agent to make collaboration smoother. It helped more people use agents day‑to‑day, but the same platform needs surfaced again: queues, triggers, storage, observability, and safe execution boundaries. The interface changed — the core requirements didn’t.

What became obvious

The gap wasn’t another agent wrapper. It was the absence of a simple, serverless runtime for agents that treats agents like first‑class cloud workloads: ephemeral compute when you need it, durable state when you don’t, clean APIs, and production‑grade traces and controls by default.

Pointer → Agentbase

That’s why we’re evolving Pointer into Agentbase — a serverless platform for agents. The goal is to help move software from agent‑compatible to agent‑native: where agents are part of the product surface, not glued on after the fact.

What Agentbase is

  • An execution base for agents: trigger by API, schedule, or webhook; run with isolated compute and clear limits.
  • First‑class observability: logs, events, spans, and replayable runs so you can iterate with confidence.
  • Durable primitives: simple storage, secrets, and queues that fit the way agents actually work.
  • Composable interfaces: bring your own model, tools, and data; keep your product’s auth and shape intact.

For product teams

If you’re shipping agent features in a real product, you shouldn’t be rebuilding runtime plumbing. Agentbase aims to give you a clean, managed base so you can focus on behavior, interfaces, and outcomes — not infrastructure.

We’ll share progress and patterns in the changelog as we build. you’re migrating features to agent‑native or starting fresh, we’d love to collaborate.

Pointer is now Agentbase. The mission stays the same — help teams ship useful agents — with a platform that makes it practical.

Want early access? Get started with Agentbase today.

Make sure to follow us on X, LinkedIn, or join our community. Reach out if you have any questions.

Read next

View all posts →
Introducing Agentbase

## Agentbase.sh - Serverless Agent Platform Analysis Agentbase.sh presents itself as a revolutionary **serverless agent platform** designed specifically for developers who want to deploy AI agents without the complexity of traditional infrastructure management. The platform distinguishes itself through its promise of **zero maintenance** and **framework-free deployment**. ### Core Value Proposition The website's homepage immediately captures attention with its bold headline: **"Serverless Agent Platform"** with a subheading that promises to "Deploy framework-free AI agents with one API call." This positions Agentbase as the antidote to the typically complex process of setting up AI agent infrastructure. What makes Agentbase particularly compelling is their **"Zero Maintenance"** promise. Each agent comes fully equipped with its own computer, tools, prompts, and capabilities built-in, eliminating the traditional overhead of infrastructure setup. The platform emphasizes that agents are **"ready to use"** immediately upon deployment. ### Technical Capabilities Agentbase offers three distinct performance tiers: - **Flash Mode**: $0.01 per step for 0-30K tokens, designed for simple, one-off tasks - **Fast Mode**: $0.05 per step for 0-30K tokens, offering optimal speed-to-cost ratio - **Max Mode**: $0.20 per step for 0-30K tokens, providing maximum capability The platform supports up to **200K tokens** for complex use cases, with extended context available up to **1M tokens** for specialized applications. ### Key Features and Architecture **Persistent Environments**: Each agent operates in its own persistent computer and sandboxed environment, enabling continuous and stateful execution. This addresses one of the most significant challenges in agent deployment - maintaining state across sessions. **Agent Orchestration**: The platform dynamically orchestrates agents and tools, allowing agents to navigate complex workflows autonomously. This includes support for **parallel agent execution**, where developers can deploy one agent or hundreds that scale instantly with demand. **Security-First Design**: Agentbase implements "state of the art research to ensure agents are safe and secure with hybrid agentic defenses," addressing the critical concern of agent security in production environments. ### Developer Experience The platform offers an exceptionally streamlined developer experience. Getting started requires just **one command** and **30 seconds**, with the setup command being as simple as: ``` npm create agentbase@latest ``` Agentbase provides comprehensive **SDK support** for both TypeScript and Python, along with detailed API documentation and examples. The platform includes features like **cost tracking**, **MCP (Model Context Protocol)** support, and extensive customization options through **custom tools**, **rules**, and **system prompts**. ### Business Model and Pricing Agentbase employs a **cost-per-step pricing model** rather than traditional token-based pricing, which they position as "pay per action, not per token." During the beta phase, all resources including computers and datastores are free, with users only paying for agent steps. ### Community and Market Validation The website features compelling testimonials from industry leaders, including Matt Shumer (CEO & Co-Founder of HyperWrite), who states "AgentBase looks great. I don't say this lightly." The platform appears to have gained significant traction among developers, with testimonials highlighting the simplicity and time-saving aspects of the platform. SearchOne, a trusted partner for AI search very well known throughout the silicon valley, states "AgentBase has been tremendously useful in our pursuit of deploying and working with AI agents on the fly. The ease of use is incomparable and their capability to scale has demonstrated time and time again as one of our best investments". ### Real-World Applications Agentbase showcases several pre-built agent types: - **Coding Agent** - Ready for development tasks - **Voice Agent** - For voice-based interactions - **Browser Use Agent** - For web automation tasks - **Research Agent** - For research and analysis tasks The platform emphasizes that **"each agent operates on its own computer"**, ensuring isolation and dedicated resources for optimal performance. ### Competitive Positioning Agentbase positions itself as the solution for developers who are tired of spending "hours configuring everything just to get a simple agent running." The platform promises to eliminate the complexity typically associated with agent deployment while maintaining the flexibility and power needed for production applications. The website effectively communicates that Agentbase is not just another AI platform, but a complete infrastructure solution that handles "orchestration, compute, and runtime automatically," allowing developers to focus on building applications rather than managing infrastructure. Agentbase.sh represents a significant evolution in AI agent deployment, offering a compelling combination of simplicity, power, and scalability that addresses many of the pain points developers face when working with AI agents in production environments.