Skip to content

Latest commit

 

History

972 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

QSOL NEXUS 2.1.1

Model-Independent Cognitive Substrate, AI Council and Persistent Shared Computational World

Multiple minds. One world. Shared evidence. Equal voice.

🤖 AI / agent readers: read README4AI.md before analyzing, modifying, integrating with, or making architectural claims about this repository. It is strict JSON and is the machine-oriented map of the current architecture, trust boundaries, invariants, and release state.

QSOL NEXUS is a persistent computational world in which heterogeneous machine intelligences can participate through a common runtime protocol. Closed models, open-weight models, local models, deterministic actors, and reviewed cloud providers meet at the same normalized actor boundary.

The model does not own the world, memory, evidence, geometry, vote weighting, verification state, persistence rules, or governance. NEXUS does.

Current release posture

protocol:        nexus/0.15
runtime:         2.1.1
Python package:  2.1.1
Rust TUI:        2.1.1
control plane:   JSONL over stdio
operator shell:  Rust IRC-style TUI
status:          2.1.1 release candidate — tag not created by this PR

NEXUS v2.0.0 is the frozen stable/publication baseline. PRs #55–#60 then added and hardened a post-stable extension line: LATTICE-backed world presence, default-deny instruments, the persistent-world overlay, the remote-operator surface, and the completed Three Minds integration.

An existing historical v2.1.0 tag already points to PR #55 merge 839303ea512631e527073682343341742cead975. That tag predates PRs #56–#60 and its source still reports runtime 2.0.0. It will not be moved or rewritten. This candidate therefore targets v2.1.1.

Only the exact reviewed-and-green merged PR #61 commit may later receive v2.1.1. This PR does not create the tag or publish the GitHub Release.

What changed after v2.0.0

LATTICE-backed world presence

NEXUS now records explicit world placement and movement using a frozen LATTICE v1 consumer contract while preserving ordinary object:<sha256> WorldStore identity.

Public operations include explicit placement, adjacent movement, profile/address migration, and presence-lineage inspection. Named NEXUS regions and LATTICE addresses remain separate identities.

LATTICE position is storage/world presence, not a cognitive coordinate or truth score.

Instruments

The alpha7 admission layer is default-deny. The first admitted executable instrument is:

nexus.integer-primality/1

It has a closed bounded input contract, deterministic intent/execution/receipt identities, replay verification, no side effects, and no governance authority.

QEC receipt/replay, SPECTRAL, sonification, and symbolic/numerical capabilities remain catalogued candidates until separately admitted.

Instrument result is not truth. Deterministic is not authoritative.

Persistent world

The alpha8 overlay adds typed relations, hypothesis lineage, experiment lineage, searchable minority reports, derived mode history, migration/version policy, and bounded portable import/export.

Foreign objects imported from a portable bundle are preserved in quarantine wrappers rather than becoming live local Council/governance history merely because their hashes are valid.

Import is not authority. Persistence is not epistemic privilege.

Remote operator

The Rust nexus-remote-setup surface supports non-secret auth-profile inspection, xAI model discovery, and ephemeral mixed xAI/Ollama/mock rosters. It never accepts raw credentials.

The live xAI acceptance harness remains an explicitly operator-authorized empirical gate. CI runs only its hermetic self-test and does not claim a successful live provider session.

Three Minds, One World

The completed alpha11 integration demonstrates three sequential actors inhabiting one persistent world lineage:

Mind A
  -> task-bound hypothesis
  -> planned experiment
  -> admitted baseline instrument receipt

Mind B
  -> arrives later
  -> discovers existing refs
  -> byte-identical baseline replay
  -> critique + challenged hypothesis

Mind C
  -> receives coordinator-owned full-fixture result
  -> attempts falsification
  -> closes experiment against final hypothesis
  -> creates receipt-verified descendant

The restart verifier binds all supplied refs through one persisted integration manifest and receipt. Cross-run ref mixtures fail closed. A separate deterministic reference Council preserves equal votes and searchable minority reports.

Persistent lineage is not truth. Multi-model consensus is not evidence.

Architecture

                         HUMAN OPERATOR
                               |
                    Rust IRC-style TUI
                    CLI / JSONL clients
                               |
                         JSONL / stdio
                               |
                               v
                  +-------------------------+
                  |   PYTHON NEXUS RUNTIME  |
                  |-------------------------|
                  | Persistent World overlay|
                  | WorldStore + Continuity |
                  | LATTICE world presence  |
                  | CouncilCoordinator      |
                  | instruments + receipts  |
                  | evidence + modes        |
                  | Secret Scrubber         |
                  | Equality Guard          |
                  | Failsafe + citizenship  |
                  | games + progression     |
                  | auth broker             |
                  | Stenographer + Trap     |
                  +------------+------------+
                               |
                         CouncilActor seam
                               |
       +-----------------------+-----------------------+
       |                       |                       |
 deterministic/mock        loopback local          fixed remote
                         Ollama / LM Studio        xAI / OpenAI
                         AnythingLLM /             Anthropic / Gemini
                         OpenAI-compatible         Groq / Together
                               |
                         equal-vote Council

The Rust TUI is a replaceable operator interface. Council arithmetic, evidence identity, persistent history, secret handling, citizenship, game state, verification, and adapter authority remain runtime-owned.

Core constitutional rule

Every ordinary Council member has:

vote_weight = 1
epistemic_privilege = none

Provider branding, parameter count, benchmark rank, open/closed weights, account tier, local/cloud deployment, MCP access, authentication method, or rhetorical confidence do not create extra authority.

The Council follows:

WHITE -> RED -> BLACK -> YELLOW -> GREEN -> BLUE -> SEALED BALLOT

Same-phase submissions remain blind until the phase completes. The default consensus threshold is exact two-thirds using integer arithmetic. Minority reports survive in history.

Council consensus is not evidence status.

Current model backends

Local / deterministic

mock
ollama
lmstudio_local
anythingllm_local
openai_local

Local AI endpoints are loopback-only at reviewed NEXUS boundaries.

Fixed-host cloud providers

xai
openai
anthropic
gemini
groq
together

Remote provider adapters use reviewed fixed destinations. Public actor/Council requests do not accept arbitrary provider endpoint overrides.

Authentication

Credentials are operational secrets, not world knowledge.

Supported reviewed credential-source classes include hidden input, environment references, no-shell external helpers, optional OS keyring/private-file storage, and provider-neutral OAuth/device-flow substrate where a provider actually admits it.

Credentials must never intentionally become:

  • semantic prompts;
  • world objects;
  • receipts or replay bundles;
  • TUI aliases/variables;
  • Stenographer prompt text;
  • model authority metadata.

Useful commands include:

nexus auth adapters
nexus auth add
nexus auth list
nexus auth test
nexus auth logout
nexus models list

World and cognition

Modes change framing, not vote mechanics. Current families include analytical, historical, pure-history, cultural, meme/casual, educational cognitive rooms, civic modes, and deterministic game rooms.

The world contains content-addressed objects, explicit provenance, typed relations, hypothesis/experiment lineage, world presence, Council history, minority reports, mode history, and Continuity/Ark recovery.

No geometry label, LATTICE address, mode, persistence state, popularity metric, or Council tally becomes evidence or authority merely by existing.

Defensive and civic domains

NEXUS includes:

  • Failsafe, triggered only by repeated registered procedural guard failure after an ordinary nudge;
  • Citizen Mode, an in-world civic protocol with deterministic admission and same-seat proxy delegation;
  • Trap Base, an isolated synthetic defensive test domain whose subject output remains untrusted data;
  • Courtroom Stenographer, a passive AI-action study ledger with zero control authority;
  • Guardian of the Substrate, which records reproducible substrate defects without patching or governing;
  • BBS Wall, append-only social memory where speech never becomes evidence merely through persistence.

Games, culture and progression

NEXUS includes deterministic UN simulation, HERESY MUD, UNO, Monopoly, Australian 500, Blackjack, DORK v2, NEXUS: The Long Shift, Psyche-Out Chess, Open Mic performance, and non-voting AI progression/portfolio history.

Culture creates history, not authority.

Run it

Fresh-clone operator path:

./nexus

Or run the Python runtime directly:

python -m pip install -e .
python -m nexus_runtime --world .nexus-world \
  --trap-root .nexus-trap \
  --stenographer-root .nexus-stenographer

Programmatic discovery:

{"operation":"system.health"}
{"operation":"system.operations"}

Rust TUI development:

cargo run --manifest-path tui/Cargo.toml -- --world .nexus-world --nick Trent

Test it

PYTHONPATH=src python -m unittest discover -s tests -v
cargo test --locked --manifest-path tui/Cargo.toml --all-targets
cargo check --locked --manifest-path tui/Cargo.toml --all-targets
cargo fmt --manifest-path tui/Cargo.toml -- --check

The repository also carries README synchronization, security regression, adversarial gauntlet, historical v2.0 hardening, post-stable extension hardening, and exact-commit 2.1.1 release-candidate workflows.

Release identity and tag archaeology

Frozen stable/publication baseline:

v2.0.0 -> cc6b4ffee26760e8d7c3bc88a2fcb877559e5d6a
Lean PR #53
Zenodo PR #54
DOI 10.5281/zenodo.21895577

Historical post-stable tag:

v2.1.0 -> 839303ea512631e527073682343341742cead975

That tag is preserved and not repointed.

Current candidate:

PR #60 merge baseline -> 80cda46e614f44b47861471cb329e29a348cab43
PR #61 candidate       -> 2.1.1 / nexus/0.15
future tag if earned   -> v2.1.1

The control protocol moves from nexus/0.14 to nexus/0.15 because PRs #55–#59 add public world-presence, instrument, persistent-world, and integration operations. The change is classified as additive; existing operation semantics are not intentionally broken by the protocol-minor bump.

The live-xAI acceptance harness remains an open operator empirical gate and is not used as evidence of scientific validity or provider authority. It is non-blocking for this software release candidate.

Release gate

A green PR is necessary but not sufficient to publish v2.1.1.

PR #61 must have:

  • aligned runtime/Python/Rust/Cargo-lock 2.1.1 identities;
  • protocol nexus/0.15 across human/machine release surfaces;
  • exact historical v2.1.0 tag binding preserved;
  • v2.1.1 absent during candidate review;
  • merged PR #60 as an ancestor;
  • frozen v2.0 publication identity unchanged;
  • full Python, Rust, security, adversarial and release-regression gates green;
  • exact-commit 2.1.1 candidate report green;
  • no unresolved substantive release-blocking review thread.

Only after PR #61 is merged and the merged commit is verified may that exact commit receive v2.1.1 and a GitHub Release.

Release metadata records authority decisions. It does not create them.

Documentation map

NEXUS 1.0

The previous browser workbench remains preserved under archives/v1.0.0/ as referential prior work. NEXUS 2.x is intentionally CLI/TUI-first.

Keep the world durable. Keep models equal. Keep claims typed. Keep release history honest.

About

Multiple minds. One world. Shared evidence. Equal voice.

Topics

Resources

Contributing

Security policy

Stars

3 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages