run your agent.
it can't destroy your machine.
a sandbox for ai agents. the agent thinks it has full access to your system. it gets a fake home. a clean environment. a network policy. an mcp scanner. every action logged. you sleep fine.
what it does
env sanitizer
strips every secret from the environment. tokens, api keys, credentials, passwords. 30+ patterns. the agent sees a clean shell. your secrets stay yours.
fs overlay
agent gets a fake home directory. ~/.ssh invisible. ~/.aws invisible. ~/.env invisible. writes go to the overlay. reads from allowlisted paths only.
net policy
allow/deny rules per host. wildcard support. the agent can't call your production database. the agent can't exfiltrate data. every connection logged.
mcp gate
every mcp server connection goes through the scanner. tool descriptions checked for prompt injection. injection patterns stripped before the agent sees them.
inject scanner
30 patterns. prompt override, identity swap, exfiltration, destructive commands, reverse shells, tool poisoning, base64 obfuscation, privilege escalation. all detected. all stripped.
audit log
sqlite. every sandbox action, every file access, every network request, every injection attempt. timestamped. queryable. you know what happened.
http api
create sandboxes, query audit logs, kill processes, manage rules. all from a single endpoint. one binary. no dependencies.
7
modules
62
tests
30
inject patterns
0
dependencies
install
go install github.com/valtors/vault/cmd/vault@latest
use
vault run -- claude-code vault run -timeout 60 -- python script.py vault serve -port 9090