Versioned technical documentation
Documentation
Architecture, runtime, model lifecycle, Teleodynamic AI, Rust-oriented design, testing, and enterprise governance. Status and verification labels distinguish research from implemented capabilities.
Research
Foundations
Documentation section for foundations.
Research
Runtime architecture
Documentation section for runtime architecture.
Research
Model lifecycle
Documentation section for model lifecycle.
Research
Rust-oriented architecture
Documentation section for rust-oriented architecture.
Research
Testing and quality
Documentation section for testing and quality.
Research
Enterprise
Documentation section for enterprise.
Research
Research tracks
Documentation section for research tracks.
- Research track overview
- Guide site versus implementation projects
- Teleodynamic AI as resource-bounded learning
- Two-timescale control loops
- No-op behavior and emergent halt
- TinyLM-16M reference workload
- Custom model container as a proposal
- Transformer runtime components
- Standard ML runtime case study
- SML terminology and acronym safety
- GGUF browser implementation boundary
- Research claim maturity
Research
Project operations
Documentation section for project operations.
- Project identity and domain map
- Canonical GGUF naming and redirect migration
- Content intake and source review
- UAI project memory
- Agent File Handoff
- Versioning and release notes
- Starter content update policy
- External implementation links
- Editorial evidence lifecycle
- Deployment and rollback boundaries
- Handoff checklist
Research
Teleodynamic AI
Documentation section for teleodynamic ai.
- Teleodynamic AI overview
- What Teleodynamic Learning is not
- Thermodynamic, morphodynamic, and teleodynamic layers
- The five commitments
- State space and system tuple
- Inner dynamics
- Outer dynamics
- Endogenous resource semantics
- Energy dynamics and boundedness
- Local teleodynamic objective
- Noop action
- Emergent structural halt
- Emergent-halt proof and hard-cap confound
- Phase structure and diagnostics
- Beneficiary and constraint closure
- Path dependence and local rationality
- Distinction Engine DE11
- Forms, genesis, and wedge actions
- Natural-gradient parameter updates
- Tropical selection
- Coalgebraic step function
- ReEntry and structural compression
- Browser resource mapping
- Teleodynamic observability schema
- Experiment and falsification protocol
- Experimental regimes and ablation design
- Limitations and open questions
- Safety, ethics, and terminology boundaries
- Teleodynamic AI and small language models
Research
Modular Tiny Model Architecture
Documentation section for modular tiny model architecture.
- Modular tiny-model architecture overview
- Installed, active, and parallel model counts
- Model scale and resource envelopes
- Skill role taxonomy
- Skill package manifest
- Independent specialist topology
- Shared base and adapter topology
- Hybrid specialist mesh
- Routing and active-set control
- Pipelines, cascades, and committees
- Teleodynamic structural operators for modules
- Dependency and compatibility graph
- Resource estimation and uncertainty
- Security, licensing, and provenance
- GGUF implementation handoff
Experimental
GGUF implementation
Documentation section for gguf implementation.
- Source snapshot and authority
- Workspace and crate topology
- Zero-dependency runtime policy
- Build profile and Visual Studio integration
- Browser application shell
- Browser boot and model selection
- JavaScript-to-WASM memory transfer
- Raw WebAssembly ABI
- Error-code contract
- Runtime state machine
- SLM v1 container overview
- SLM fixed header layout
- Tensor directory and alignment
- Tensor naming and shape contracts
- Container checksum and integrity limits
- BTOK byte tokenizer
- BPE1 tokenizer
- Model loader and tensor storage
- F32 tensor path
- Q8_0 storage and matvec
- Q4_0 storage and matvec
- Tied output projection
- Transformer forward pass
- RMSNorm and residual flow
- Rotary positional embedding
- Causal attention
- KV cache
- SwiGLU feed-forward network
- Generation prefill and decode
- Sampling and deterministic PRNG
- Reusable scratch memory and logits
- Runtime diagnostics JSON
- Model artifact catalog
- Provenance manifests
- Trained-source conversion
- Native eval runner
- Quality gates
- Local static server
- Rust mini browser harness
- Test matrix and evidence
- Security and trust boundaries
- Current performance model and bottlenecks
- Current implementation limitations
- Path to modular tiny models
- Path to a Teleodynamic controller
- Release and rollback
- Browser-side provenance rendering
- Browser transcript and context semantics
- Model-load and generation timing
- Model format versus project name
- Implementation memory handoff gap
Experimental
Implementation operations
Documentation section for implementation operations.
- Runtime ownership and global lock
- ABI buffer ownership and 128 MiB transfer ceiling
- Model-load transaction and rollback semantics
- Generation transaction and recovery boundary
- Reset context, clear transcript, and free model
- Prompt token boundary and EOS handling
- Forward scratch sizing and reuse
- KV cache sizing, indexing, and commit discipline
- Storage-dispatched tensor operations
- Q8 row scales and Q4 block scales
- Sampler validation and fixed candidate storage
- Deterministic PRNG and replay limits
- Diagnostics schema and measurement ownership
- Browser boot state and stale-load suppression
- Browser artifact selector contract
- Provenance sidecar parsing and trust gap
- Transcript state versus model context
- Local server request and path safety contract
- Browser harness contract and blind spots
- SLM packer command surface
- Raw trained-source admission
- Runtime-smoke promotion gate
- Assistant-quality promotion gate
- Release build profile and artifact placement
- Visual Studio and PowerShell wrapper contract
- Unsafe boundary inventory
- Model-load failure runbook
- Generation failure runbook
- Worker-owned runtime migration
- Verified persistent model store
- GGUF compatibility implementation path
- Modular skill package ABI from the current runtime
- Teleodynamic controller integration points
- Operational compatibility matrix
- Implementation benchmark record schema
- Fuzz and property-test plan for current parsers
- Release rollback and cache invalidation
- Source-to-site evidence synchronization
- Exact implementation source map