BORDZERO

by Bord Labs

Bord Zero is a research-first harness for long-running AI agents. It provides working memory, structured state, observability, and eval-backed iteration for agents that need to work for hours, not minutes.

Join Preview
The Problem

Most agents look good
in a demo. Real work
is different.

Long-running tasks accumulate context, tool calls, and partial state. They cross sessions, branch unexpectedly, and fail in ways that are hard to debug. Most agent systems collapse under this pressure.

Current agent frameworks optimize for the first five minutes of execution. Real work lasts hours, crosses multiple tools, accumulates state across sessions, and breaks in ways that are hard to inspect. Bord Zero is the harness built for that reality.

Core Pillars

Working Memory

Bord Zero replaces transcript accumulation with structured working memory. Agents maintain key facts, decisions, and artifacts across context windows, reducing overload and improving continuity over long tasks.

Observability

Bord Zero makes every agent run fully inspectable. Operators trace each step, artifact, and failure with full provenance. No black boxes in production.

Evals

Bord Zero measures agent improvement through structured evals, not anecdotes. Eval-backed iteration is a first-class concern, built into the harness rather than bolted on after deployment.

Safe Autonomy

Bord Zero enforces boundaries, checkpoints, and clear operator control for long-running agents. Agents operate within defined safety constraints at every step of execution.

Current Focus

Starting with software
and operational workflows.

Bord Zero targets software development and operational workflows first. The goal is not just better code generation — it's better execution over time through working memory, artifact tracing, and measured iteration.

If the harness architecture generalizes, it can extend into research, operations, and hardware-heavy workflows. But Bord Labs is leading with depth, not breadth. One domain, done well, before expanding.

Research Agenda

Open questions
we're working on

Long-running execution

Bord Zero maintains agent coherence across many context windows and sessions without degradation.

Working memory & context engineering

Bord Zero manages structured state that goes beyond raw transcript accumulation in long tasks.

Artifact-based handoffs

Bord Zero passes structured outputs between sessions instead of dumping raw context into new windows.

Eval-driven improvement

Bord Zero measures harness performance systematically through evals, not through anecdotal demo results.

Dynamic internal workflows

Bord Zero supports flexible agent behavior on top of stable contracts and interfaces.

Multi-agent coordination

Bord Zero adds multi-agent patterns only when they measurably outperform a single well-harnessed agent.

Principles

Start with a strong harness, not a sprawling platform.

Keep the substrate stable; let behavior become dynamic.

Prefer structured state over prompt sprawl.

Use multi-agent patterns only when they measurably help.

Treat observability and evals as part of the product, not just infrastructure.

FAQ
What is Bord Zero?
Bord Zero is a research-first harness for long-running AI agents, built by Bord Labs. It provides structured working memory, run observability, eval-backed iteration, and safe autonomy for agents that need to work across hours-long tasks, multiple sessions, and complex tool chains.
What is an agent harness?
An agent harness is the infrastructure layer that wraps an AI agent, managing its memory, tool access, execution state, and observability. Unlike orchestration frameworks that coordinate multiple agents, a harness focuses on making a single agent reliable over long-running tasks by providing structured state, checkpoints, and inspectable execution traces.
How is Bord Zero different from other agent frameworks?
Bord Zero focuses on long-running agent execution rather than demo-scale tasks. It replaces transcript-based context with structured working memory, makes every run inspectable through full observability, uses evals to measure improvement systematically, and enforces safe autonomy with clear operator boundaries and checkpoints.
What is working memory for AI agents?
Working memory for AI agents is structured state that persists across context windows and sessions, replacing raw transcript accumulation. Instead of feeding an agent its entire conversation history, working memory stores key facts, decisions, and artifacts in a structured format that reduces context overload and improves continuity.
What does context engineering mean for agents?
Context engineering is the practice of carefully managing what information an AI agent has access to at each step of execution. This includes deciding what enters the context window, how state is summarized across sessions, how artifacts are handed off, and how to prevent context overload. Bord Zero treats context engineering as a core infrastructure concern.
When will Bord Zero be available?
Bord Zero is currently in research and development. Bord Labs publishes architecture notes, eval results, and early product updates to the research preview list. Sign up at bordzero.com to follow progress and get early access.

Follow the research

Get updates as Bord Labs publishes architecture notes, eval results, and early Bord Zero releases.

Please use your company email. No spam — architecture notes and release updates only.