Introduction to Exo
Exo is an open-source agent harness built for recursive self improvement.
$ open "Meet exo"
Exo is an open-source agent harness built so an agent can rebuild itself safely. Launch a real exo session in the terminal and have your...
checkpoints run as you work
✓ checkpoint passed, lab 2 unlocked
What you'll build
Curriculum
6 labs, about 58 min
- 1
Meet exo
Exo is an open-source agent harness built so an agent can rebuild itself safely. Launch a real exo session in the terminal and have your first conversation.
8 min
Pro - 2
Bindings and secrets
Exo keeps credentials in secrets and configuration in bindings, so the agent can rewrite its own policy without ever touching a key. Register a second model binding and switch a conversation onto it.
9 min
Pro - 3
Give your agent a shell
An environment belongs to a conversation, not to every agent by default. Create an agent backed by a sandbox provider and watch it do real work through its shell tool.
10 min
Pro - 4
Read the event log
Exo records every turn as append-only events in a layer the agent cannot rewrite. Dump the raw log behind your conversation, filter it by type, and resume the thread it belongs to.
9 min
Pro - 5
Fork the conversation and travel back
An agent's whole state is the version of its event log, so any point in that log is a branch point. Fork a conversation, take the branch somewhere new, and see both histories side by side.
10 min
Pro - 6
Capstone, run a real experiment
Put the log, the fork, and the shell to work. Branch one conversation two ways, send a different prompt down each branch, and use the event log to measure which one worked better.
12 min
Pro
Your instructor
Elvis Saravia
Founder, DAIR.AI
Elvis founded DAIR.AI and wrote the Prompt Engineering Guide, one of the most widely used references on working with language models. He has worked at companies like Meta AI and Elastic, and specializes in AI agents, RAG, and harness engineering.
How labs work
Live workspace
You run the real tools in a browser workspace. Nothing to install.
Checkpoints
Each lab checks your files and output, then unlocks the next one.
Your work stays
The workspace persists between labs so you finish with a project.