Documentation
The work, and what supports it.
Start with a product’s purpose and current scope, then explore the public demonstrations. For detailed technical evidence or a proposed integration, contact the lab.
Choose a starting point
Nibiru Engine
How encrypted execution and semantic receipts fit together.
Anansi demo
A guided simulation, a session specimen, and a retained review case.
Lab philosophy
The connection between theory, experiments, and engineering.
Machine-readable pages
Markdown mirrors and the read-only site index. No public transactional API.
Computer architecture research
PaleFox
Develop a programmable ternary computer architecture, from formal semantics to hardware.
Current scope: Defined semantic results are Proved. Selected hardware subsets are Implemented + Verified finite + Synthesized. Complete-system FPGA and ASIC execution remain Open.
Ternary inference
SandCat
Develop a software inference engine for loading and running packed ternary models, building on PaleFox representations and numeric contracts.
Current scope: Specified implementation plan: Apple Silicon first, then NVIDIA CUDA. Engine implementation has not started. No SandCat performance result or model qualification is claimed.
Learning and contextual programming
Structural Descent
Study learning directly in executable structure through local changes under explicit objectives.
Current scope: Defined local-search results are Proved and machine-checked. Experiments retain their own finite scope. Local terminality does not establish global optimality. General runtime refinement and broader model-quality claims remain Open.
Confidential computation
Nibiru Engine
Compute on encrypted data and verify a receipt for a declared execution.
Current scope: Implemented + Verified finite for recorded execution and receipt profiles. Core bounded-computation theorems are machine-checked. Broader scale, remaining cryptographic bindings, and production qualification remain Open.
Governed context and control
Baobab
Declare goals, permissions, required evidence, budgets, and rules for accepting agent progress.
Current scope: Implemented + Verified finite for recorded serial SDK profiles. Conformance is to a declared policy and verifier. Independent release qualification and control of tools outside those profiles remain Open.
Agent execution and code review
Anansi
Join an exact action or code change to its intent, permissions, execution evidence, and review decision.
Current scope: Implemented + Verified finite for recorded review and execution flows. The recent local private-action campaign uses test doubles for identity and planning. Complete integration and independent release qualification remain Open.
Eligibility and agent credentials
Heimdall
Check defined eligibility and Know-Your-Agent mandate statements while limiting disclosure.
Current scope: Implemented + Verified finite for the standalone package and recorded credential profiles. Regulated KYC sufficiency, revocation, unlinkability, and conduct-within-policy remain Open.
Model and agent orchestration
Esu
Assign organization-scoped identities from verified Heimdall presentations and route signed evidence between tasks.
Current scope: The identity core is Implemented + Verified finite. A complete standalone orchestration package remains Open. Identity and signatures do not grant task authority or establish semantic truth.
Receipt coordination research
Tiamat
Explore settlement, agent micropayments, and compute credits around execution receipts.
Current scope: Conjectural / Open. Deterministic simulations treat receipt validity as an abstract input. They do not establish consensus security, monetary safety, or a live settlement network.
How to read the evidence
Each label applies to a stated claim and scope. No label covers a complete product unless the source explicitly establishes that scope.
- Proved
- A mathematical argument establishes the stated result under its assumptions. Where we say machine-checked, a proof assistant has checked the exact formal statement.
- Verified
- A script, test, or benchmark checked the stated cases. Verified finite preserves that finite scope.
- Implemented
- The code path exists and has run on named fixtures.
- Specified
- The design is precise enough to build; implementation may still be incomplete.
- Conjectural
- A motivating idea whose stronger claim is not yet proved.
- Open
- A requirement or research question remains unresolved.
- Unresolved
- The available evidence or verifier result does not settle the decision.
Existing product notes were reviewed on 18 September 2026. The ternary research framing and SandCat scope were reviewed on 22 September 2026. Demos preserve the scope of their own retained exports. Contact the lab for the evidence relevant to a specific workload.