LiMiT · Deterministic workspace
Deterministic tools and learned functions
HuLaHo v35 combines explicit computation with private memory and bounded learning. The default provider is deterministic. Optional Ollama remains separate and requires explicit selection.
Supported tool families
- Arithmetic, exact rational calculations, units, statistics and linear/quadratic equations.
- Private facts, source documents, relationship chains and curated public reference notes.
- Rule inference, hypothetical scenarios, backward proofs and Boolean counterexamples.
- Finite constraints, linear-objective optimization, dependency schedules and bounded goal planning.
- Supplied-model Bayesian updates and diagnostic information-gain ranking.
- Numeric/text function synthesis, typed composition, regression tests and version rollback.
- Taught invocation phrases and exact phrasebook translations.
Learn by doing
- Conversation and shared learning — Learn supported facts from chat, correct memories, teach by demonstration and contribute reusable functions.
- Teach a deterministic function — Learn numeric and text functions from examples, add phrases and inspect versions.
- Private memory and source documents — Understand explicit facts, source retrieval, workspace identity and recovery.
- Troubleshoot a chat stream — Separate browser, identity, proxy and backend failures when HuLaHo chat does not respond.
- Exact decimal and fraction calculations — Compute supported arithmetic with exact fractions and inspect decimal output limits.
- Bayesian updates and diagnostic tests — Update supplied probability models and rank tests by expected information gain.
- Finite constraint solving and optimization — Search explicit variable domains and distinguish best-found results from proven optima.
- Logic, rules and missing-premise explanations — Use explicit premises to inspect deductions, counterexamples and unresolved goals.
These capabilities have explicit size and search limits. The product does not claim AGI, arbitrary language fluency, autonomous web research or universal correctness from a few examples. Read the methods page before interpreting a result as more than the supplied model supports.