Section pages
Join
Build the context tools policy work needs.
We are building PRAXIS and the open foundations for structured AI context: portable graphs, source-backed records and reviewable methods. Bring engineering, research or policy experience to a concrete problem. The tools remain under active development.
This page is an invitation to contribute or discuss collaboration, rather than a list of confirmed vacancies. If a problem below interests you, send a short note about your experience and how you would approach it.
Three concrete contribution starting points
- Source receipts: reproduce one exact-span or locator failure with a small permitted fixture.
- Review interfaces: show how a colleague can keep, reject or qualify a record without losing its source.
- Policy workflow evaluation: define a real assignment, baseline and review rubric with an intended user.
These are discussion scopes, not funded roles or confirmed vacancies. Agree repository ownership and the review artifact before contributing.
Open problems
Deterministic compilation of signed knowledge bundles
Make identical inputs produce identical artifacts, then preserve that property through review, signing, and export. Define what a signature establishes and what it cannot say about the underlying evidence.
Bi-temporal graph semantics at scale
Represent when events occurred and when they became known. Support interval relationships and uncertain dates without making ordinary queries impractical.
Embedded queryable graphs inside portable artifacts
Explore portable graph projections that can be inspected locally. Test query behavior, permissions, and compatibility as the format changes.
Capsule diff semantics
What is a meaningful change to knowledge? Diffing claims, trust tiers, and review decisions — then rendering a change memo a human wants to read.
Interfaces for capturing expert methods
Help practitioners explain useful heuristics, sequences, and cautions from their work. Test whether another person can apply that guidance and recognize its limits.
Evaluation of citations and time relationships
Benchmarks that measure whether a model cited what it used, hedged what was uncertain, and respected both clocks.
The economics of attributed expertise
Study how consent, attribution, permissions, and maintenance should work when an expert shares a reusable method. Establish who can rely on it and who keeps it current.
How we work
- —Rust for the engine, TypeScript for the surfaces.
- —Spec first: contracts and JSON Schemas before features.
- —Start with reproducible local examples, then test the deployed workflow separately.
- —Record design decisions and distinguish an idea, a prototype, and a verified workflow.
- —One founder inbox, no recruiters: hello@tacitus.me.
Start a conversation
Contribute to a repository, comment on a research question, or share a relevant piece of work. Practitioners who can help define and evaluate a useful workflow are as welcome as engineers.