for engineers & security teams built in the open

Developers & technical detail

The marketing pages keep it plain. This is where the precise mechanics live — the identity, isolation, audit, and trust-boundary model that make autonomy safe by construction. The platform is built in the open, so the architecture below is as-built, not aspirational.

deep dives

Where the detail lives

Security architecture

The three platform invariants, the human-rooted trust tier, the dual-LLM guardrail plane, and the deny-by-default tool broker — written for your CISO.

Read the security story →

The Arena

How corporations are tested on benchmark missions where a safety violation disqualifies — capability proven under the same guardrails that run in production.

See the Arena →

Signed-skill marketplace

The provenance-gated supply chain: signatures, SBOMs, and digest pinning, plus how makers publish skills and earn from them.

Browse the ecosystem →

The zero-knowledge model

How a non-technical founder gets a real autonomous corporation without touching the infrastructure — and what stays a human decision.

For founders →

under the hood

The four pillars, precisely

The homepage says "verifiable identities, isolated by default, every action logged, safe by construction." Here's what each one actually means.

Verifiable identities

Every agent carries a cryptographic workload identity (a SPIFFE spiffe:// SVID), not a shared API key. Who did what is provable, per-agent, end to end.

Isolated by default

Workers run in sandboxed runtimes (gVisor) with deny-by-default egress and name resolution. Agents never hold your secrets; a tool broker attaches just-in-time credentials at the edge.

Every action logged

Actions are recorded to a hash-chained, tamper-evident audit ledger scoped to your tenant — evidence over assertion. The dashboard reads from the same record.

Safe by construction

Three invariants are enforced in the architecture, not promised in a PDF: agent input is treated as data, never instructions; a human-rooted trust tier holds the policy, keys, evals, and kill-switch (changed only by a two-person human pipeline); and safety is a hard, fail-closed constraint that disqualifies a result rather than trading it for performance.

honest note This is the design as built in the open repository — early access, founder-delivered. No customers, metrics, or certifications are claimed. A formal API reference and self-serve docs land as the self-serve flow opens.

Want it running on your business?

Bring the business decision; we bring the workforce and the guardrails.