Engineering

How to Build a Code Interpreter on Freestyle

A practical guide to building a code interpreter for AI agents on Freestyle VMs: sandboxed code execution, snapshotting, live forking, and suspend/resume sessions.

How to Build Deep Research Agents on Freestyle

A practical guide to building deep research agents on Freestyle VMs: snapshot findings, live-fork agent state, and run parallel exploration without redoing work.

How to Deploy Long-Running Agents on Freestyle

A practical guide to running long-running agents on Freestyle VMs: persistent sessions, suspend and resume, snapshots, live forking, and stateful microVMs for AI agents.

How to Do Reinforcement Learning on Freestyle

How to run reinforcement learning for AI agents on Freestyle VMs: cache a golden environment, fan out parallel rollouts, collect trajectories, train at scale.

How to Run AI Evaluations on Freestyle

A practical guide to running AI evaluations and agent evals on Freestyle VMs: ephemeral microVMs per test case, full traces, and burst parallelism.

Git Agents

Git agents are AI agents that use Git as their workspace, memory, and review layer. Here is how Git agents should branch, commit, diff, rollback, and ship with Freestyle Git.

Version Control for AI Agents

Version control for AI agents means branchable, reviewable, auditable state for AI-generated code, docs, assets, and workflows. Freestyle Git gives agent platforms real Git repos through an API.

8 Things You Didn't Know Git Already Did

Git is not just commit, push, and pull. It already ships with debugging, parallel workspaces, rollback, patch review, and agent-friendly history tools.

The best filesystem for AI agents is Git

The best way to handle an agent's filesystem is Git: normal files for the agent, and commits, branches, diffs, rollback, permissions, and review for your product.

Our teams' favorite Rust crates 2025

A list of our favorite Rust crates that we think all Rustaceans should know about.

The Future of AI Coding Agents

Where AI App Builders and AI Software Engineers are headed next.

The missing trust model in AI Tools

The tools we give AI are not safe, that's going to cause problems.

How we revamped our Docs for AI

What do docs look like if people don't read them?