Adaptive context engine
for AI coding agents
Equip the right context for the job.
Different work needs different context. A loadout is a reusable kit — context for your environment, a workflow for your task — and it keeps itself current. Launch your agent and the right one is equipped.
$ curl -fsSL loadout.tools/install | sh
prebuilt binary · macOS & Linux · or npx @ellery/loadout
Your AGENTS.md describes the repo. Loadout carries how you work.
The things that follow you across projects, not the repo — rendered into a local, gitignored overlay for whichever agent you launch.
Reusable context for the environment
Small, composable units of guidance — conventions, build commands, comms style — plus live probes that capture your real toolchain, hosts, and containers at render time.
Your preferred workflow for the task
Each loadout carries a way of working — brainstorm, plan, implement, verify, ship — injected as stages your agent follows and slash commands you invoke.
Self-updating, self-healing
Every launch re-renders your context against the machine as it is now. Drift gets detected and healed — and with learning on, Loadout proposes improvements mined from your own sessions.
Loadout reads the project and matches it to targets. Selection is deterministic and inspectable — no LLM decides.
$ load explain Detected targets: [rust] Loadout selection → rust
One loadout per context: a named kit of fragments, composed in TOML you own and edit in load studio.
[[loadouts]] name = "rust" targets = ["rust"] fragments = ["conventions", "voice"]
load claude renders the overlay, wires it into the agent, and starts it — context equipped.
created .loadout/generated/claude.md updated CLAUDE.local.md ▸ launch claude
WORKS WITH
Each agent reads your context from a local, gitignored file — overlays, bindings, and logs never land in the repo. On a bare server with no repo at all, the machine target has you covered.