From zero to a governed, verifiable agent — using only a stable Rust toolchain. No SaaS, no keys, fully offline.
Each step is deterministic and offline. Nothing reaches a provider until governance allows it.
Compile the compiler, the VM and the conformance runner from source.
cargo build --workspaceDeclare a module, agents, a passport and a Policy v2 block in a .argx file.
vim runtime.argxValidate semantics and verify the emitted bytecode before running anything.
argorixc verify-bytecode runtime.argxRun the VM in dry-run, then read the exported security report and evidence bundle.
argorix-vm run runtime.argbc.json --dry-runEvery concept maps to a declarative block the compiler, runtime and policy system can check.
A technical, legal and security ID: who the agent is, who provides it, its jurisdiction, risk and evidence.
Open Passportrequire / deny / on-violation rules, evaluated before any action. Fail-closed by default.
dry_run, simulated and the guarded sandboxed_external — the runtime only does what governance permits.
See Live MVPA portable, offline-checkable bundle with SHA-256 digests over canonical serialization.
Providers are non-executable by default; secrets are referenced, never read.
Named threats (e.g. prompt-injection) mapped to guards that hold turns for review.
A minimal governed agent — declarative, compiled to bytecode, checked before it runs.
module ArgorixChatbotRuntime agent AssistantAgent { receives UserPrompt capabilities { trace.write } } passport AssistantPassport { global_id = "argx:agent:01HZX9CHATASSISTANT" risk_level = high jurisdiction = "CL" } policy ChatbotPolicy { deny external.execute on-violation => review // fail-closed }
The full language reference, examples and conformance suite live in the repository.