Business ontology modeling + AI-assisted decision platform built on DeepSeek Harness (dsh).
Turn expert business criteria into machine-validatable, issuable, derivable structures — O1 objects / O2 behavior rules / O3 scenario cards. One line to remember:
Judgment lives in the rule engine; determination lives with humans.
Universal pipeline: Knowledge → Ontology → Platform assembly → Decision support. Any domain swaps package content only — the skeleton stays the same.
Chinese original: README.zh.md · docs/overview-pipeline.zh.svg
⓪ Choose before chat. The new-session hero chip pre-selects an expert preset (AML / TAX /
MED) and the session header keeps the active expert in view. A started session's composition
cannot be hot-switched — use expert_agent to delegate to an expert child agent, or open a
new session.
① Build. One sentence creates the model: create_package (with humanApproved) →
skeleton proposal → save_skeleton → sliced drafting (draft_elements; drafts enter the
queue, never the package) → human review on the canvas → materialization through the I6
skeleton gate → validate → derive → load gate → human issue (0 errors + ≥ 1 scenario card +
≥ 1 score rule; review/issue transitions re-derive in the same transaction).
② Apply. An issued package is a read-only application — run_scenario / invoke_expert,
the Web Scenarios tab, or the CLI. Every run yields candidates, risk scores, hit rule IDs,
evidence chains and pending checks with humanReviewRequired=true: machines screen clues and
assemble evidence; humans decide.
③ Operate. Each run turns the loop: run records (audit + forbidden-phrase scan) → experience signatures → recall hints (never re-scores) → human annotation → SkillTemplate drafts → approved methodology upgrades. ISSUED is read-only — evolution forks a new version; a knowledge side-path (conversation turns, uploaded docs) feeds the model through the human-approved queue.
Business judgment criteria usually scatter across code, configs, and individual experts' heads: every AI capability re-aligns the same semantics, thresholds drift between releases, and compliance boundaries get bolted on after the fact. dsh-onto makes the criteria the product: they are modeled as versioned YAML packages, validated by a rule engine, issued through human review, and derived into fingerprinted runtime contracts — while the LLM is kept on a short leash (drafting only, never scoring, never issuing verdicts).
- Packages — create / import / export / version fork / human issue (
DRAFT → IN_REVIEW → ISSUED; ISSUED is read-only) ·dsh-onto package *· Web - Conversational modeling — one sentence → skeleton → sliced drafts → human review → materialize behind the I6 skeleton gate · main chat · Web
- Knowledge pipeline — 6 source types → LLM parsing → candidate queue → human approval ·
dsh-onto knowledge *· Web - Validator & deriver — reference closure, weights,
derived/**fingerprints, load gate (DERIVED_DRIFT) ·dsh-onto validate/derive/gate - Rule engine — exclusions then weighted scoring; every score points back to a rule ID · core
- Scenario runs — pipeline dispatch, three-layer audit, run history A/B compare, audit
export ·
dsh-onto run/runs *· Web - Trusted business execution — three-state data status, six-tuple authorization,
decision→proposal ledger, external receipts + reconciliation (no receipt ⇒
unknown), ticket-sim & test adapters · lives in@dsh-onto/runtime·record_review_decision/create_action_proposal/execute_behavior_action/reconcile_action - Experience loop — run records → recall hints → human annotation → SkillTemplate
upgrades ·
dsh-onto exp *· Web - Expert presets — AML / TAX / MED picked before chat;
expert_agentdispatches expert sub-agents · hero chip · main chat - Multi-domain — AML + MED + TAX ship as package content only, zero engine changes ·
packages/
Dependency direction: core ← runtime ← backend (assembly) · workbench talks to
backend over remote services only.
dsh-onto/
core/ @dsh-onto/core — pure engine library + CLI (zero harness deps; 236 node --test acceptance tests)
runtime/ @dsh-onto/runtime — shared run & governance library: data connectors,
identity/authorization, run records, review/approval, action ledger,
reconciliation (no auto dsh assembly; backend and future business
plugins depend on it — never the reverse)
backend/ @dsh-onto/backend — dsh bundle: development wires (model/knowledge/issue)
+ assembly shell over @dsh-onto/runtime (onto* landing point; 289 tests)
workbench/ @dsh-onto/workbench — dsh bundle: Web UI (six tabs + persistent right-rail ontology graph)
mock-services/ ticket-sim — standalone ticket simulation service (self-persisting SQLite,
idempotent across processes, fault injection, reconciliation; receipts always `simulated`)
packages/ example business ontology packages (first-class deliverables): aml/1.0.0, med/1.0.0, tax/1.0.0
examples/ scenario run fixtures (aml-run.json / med-run.json / tax-run.json)
scripts/ setup-profile.ps1 / dev-web.ps1 / dump-config.ps1
ensure-expert-presets.ps1 (+ expert-presets.json, preset-plugins.standard.yaml: idempotent expert-preset block)
docs/ architecture.md — self-contained concepts & rules (Chinese)
overview-pipeline.{zh,en}.svg — end-to-end pipeline diagrams
- Node.js ≥ 24.9 and pnpm 11.x
- A deepseek-harness checkout as a sibling
directory (
../deepseek-harness), already run throughpnpm install && pnpm run build. The harness is a read-only dependency — dsh-onto never modifies it.
# 1. Create the `dsh-onto` profile, install both bundles, seed the example packages
dsh-onto\scripts\setup-profile.ps1
# 2. Assert the composed profile still contains @dsh-onto/backend + @dsh-onto/workbench
# (runtime ships as a library dependency of backend, not a separate bundle)
dsh-onto\scripts\dump-config.ps1
# 3. Launch the Web UI (open the printed dsh web URL) → sidebar entry "dsh-onto"
dsh-onto\scripts\dev-web.ps1The workbench exposes six tabs: Packages · Modeling · Knowledge · Validate & Derive · Scenarios · Experience (plus a persistent Ontology graph in the right rail).
cd dsh-onto\core
pnpm install; pnpm run build
node dist\cli.js package create --name demo --version 1.0.0 --domain DEMO --root ..\workspace
node dist\cli.js package issue --name demo --version 1.0.0 --by <human> --root ..\workspace
node dist\cli.js validate ..\packages\aml\1.0.0
node dist\cli.js derive ..\packages\aml\1.0.0
node dist\cli.js gate ..\packages\aml\1.0.0
node dist\cli.js run ..\packages\aml\1.0.0 SCN-AML-FIND-CLUES --data ..\examples\aml-run.json --out result.json
node dist\cli.js runs compare <runA> <runB> --root ..\workspace
node dist\cli.js knowledge upload --package demo --type LEGACY_SYS --name algo --file algo.txt --root ..\workspace
node dist\cli.js knowledge parse --source <id> --llm deepseek --root ..\workspace
node dist\cli.js exp recall --dir ..\packages\aml\1.0.0 --scenario SCN-AML-FIND-CLUES --data ..\examples\aml-run.json