Basecamp
DownloadGet Started
ProductSandboxesEnvironmentsStacksDocsDownloadGet Started
Docs/AI environments

AI environments

Treat AI resources as environment inputs: prepare them before the agent starts, lock what matters, and keep credentials and model execution outside the environment definition.

New to Basecamp?Start with the quickstart →

AI is an environment problem too

A coding agent may need far more than a compiler. Browser binaries, MCP servers, model files, tokenizers, GPU libraries, datasets, indexes, and utility binaries are all inputs that can make one sandbox behave differently from another. Basecamp's environment graph is intended to make these inputs explicit and reproducible instead of downloading them ad hoc during agent execution.

Resource families

AI environment resource graph around a ready Basecamp environment
AI workloads can depend on far more than language packages; the environment model makes those inputs explicit.
ready environmentbefore agent execution
runtimeRust · Node · Python
agent toolsMCP · browser · CLI
nativeCUDA · ROCm · SDKs
modelsweights · tokenizer
datadatasets · indexes
packagesnative ecosystems
FamilyExamplesv0.5 status
Language runtimesRust, Node, Python, GoRust shipping; others roadmap
Package setsCargo, npm/pnpm, uv/pip, Go modulesCargo shipping; others roadmap
AI model assetsweights, tokenizers, adapters, embeddingsEnvironment-model target
Agent toolsMCP servers, skills, browser tooling, static binariesEnvironment-model target
Native / GPUCUDA, ROCm, compiler SDKs, system librariesEnvironment-model target
Datadatasets, indexes, fixtures, test corporaEnvironment-model target

Agent-native without becoming an agent framework

Basecamp should make the environment understandable to automation: deterministic locks, machine-readable schemas, status, and doctor. The coding agent can verify a prepared environment and move directly to repository work instead of spending context and network access discovering how to bootstrap the machine.

The current boundary

Basecamp does not run the model or agent loop.

It prepares the reproducible environment in which the model, agent harness, and native project tools operate. Secrets and API credentials should remain external to basecamp.lock.