How hamdok/aestheris works
Aestheris is a local trust gateway for AI agents. Its main workflow is aestheris run: load policy and vault secrets, launch a loopback proxy, then run the agent in a sandbox with phantom credentials; proxy requests are checked, privacy-filtered, forwarded, and audited, with sensitive actions optionally awaiting human approval. Separate CLI capabilities initialize configuration, scan for exposure, and inspect audit evidence. The tree and sampled excerpts support the principal wiring; some CLI dispatch and subsystem details are documented but not fully sampled.
Main entry points: proxy.rs Local gateway · run.rs Run coordinator · main.rs CLI dispatch · audit.rs Audit log
Components 20
CLI workflows
- CLI dispatch
src/main.rs - Initialize setup
src/init.rs - Exposure scan
src/scan.rs - Project findings
src/project_scan.rs - Secret detection
src/secret.rs - Gitleaks rules
src/gitleaks.rs
Gateway controls
- Run coordinator
src/run.rs - Policy rules
src/policy.rs - Local gateway
src/proxy.rs - Agent sandbox
src/sandbox.rs - Linux isolation
src/sandbox_linux.rs - Human approval
src/approval.rs
Privacy and secrets
- Encrypted vault
src/vault.rs - Phantom tokens
src/phantom.rs - Privacy shield
src/privacy.rs
Evidence and review
- Audit log
src/audit.rs - Audit reports
src/report.rs
Other components
- Developer
- AI agent
- API / model provider
Connections 27
- Developer → CLI dispatch: runs command
- CLI dispatch → Initialize setup: initializes
- CLI dispatch → Exposure scan: starts scan
- Exposure scan → Project findings: checks project
- Project findings → Secret detection: detects secrets
- Project findings → Gitleaks rules: uses rules
- CLI dispatch → Run coordinator: starts session
- Run coordinator → Policy rules: loads policy
- Run coordinator → Encrypted vault: unlocks secrets
- Run coordinator → Local gateway: starts gateway
- Run coordinator → Agent sandbox: launches agent
- Agent sandbox → Linux isolation: uses Linux isolation
- Run coordinator → Human approval: opens approval channel
- Run coordinator → Audit log: records session
- Run coordinator → AI agent: launches with phantom credentials
- AI agent → Local gateway: sends requests
- Local gateway → Policy rules: checks route rules
- Local gateway → Phantom tokens: validates token
- Local gateway → Privacy shield: filters payloads
- Local gateway → Human approval: requests decision
- Local gateway → API / model provider: forwards request
- API / model provider → Local gateway: returns response
- Local gateway → Audit log: records requests
- CLI dispatch → Human approval: handles approval command
- CLI dispatch → Audit log: reads audit events
- CLI dispatch → Audit reports: renders report
- Audit log → Audit reports: provides events