Feature Parity Matrix

Which surface (SDK, CLI, backend API, dashboard, docs) exposes each MultiMind feature, and the remaining gaps.

Audited and standardised 2026-10-12 against version 1.0.0. The rule this page tracks: a feature that works in the CLI must also be reachable from multimind backend, visible in the governance dashboard where a UI makes sense, and documented here. A dash means the surface does not exist. CI runs scripts/check_docs_parity.py, which fails when a CLI command or a backend route is not named anywhere under docs/.

The API column lists routes on multimind backend (see openapi-platform.json). The dashboard has eight views: overview, audit, costs, inventory, guardrails, scan text, evidence report, compliance.

Matrix

Feature	SDK module	CLI	Backend API	Dashboard	Docs	Status
PII scan / redact	compliance/guard.py	compliance scan-text	POST /api/scan-text	Scan text view	CLI, dashboard	Stable
Guard wrapper guard(model)	compliance/guard.py	— (SDK)	— (via proxy)	—	compliance, integrations	Stable
OpenAI-compatible guard proxy	gateway/guard_proxy.py	serve	/v1/chat/completions, /v1/embeddings, /v1/guardrails/reload	Reload button in Guardrails view	guard proxy	Stable
Guardrails file	dashboard/server.py	serve --config	GET/PUT /api/guardrails	Guardrails view	dashboard	Stable
PII audit trail	compliance/guard.py AuditLog	—	GET /api/audit, GET /api/summary	Audit, Overview	dashboard	Stable
Cost tracking, budgets, chargeback	observability/cost_tracker.py	audit costs --period	GET /api/costs?period	Costs, Overview	AI inventory	Stable
Shadow-AI inventory	observability/ai_inventory.py	audit scan --include-env	GET /api/inventory?include_env	Inventory	AI inventory	Stable
Evidence report	compliance/reporting.py	compliance report-evidence	GET /api/evidence-report?format&period&organization&project&costs_log	Evidence report view	compliance, dashboard	Stable
Compliance monitor and report	gateway/compliance_api.py	compliance run-compliance, generate-report	POST /v1/compliance/monitor, /report, GET /regulations, /healthcare/use-cases	Compliance view (record a check)	compliance	Stable
Compliance dashboard and alerts	gateway/compliance_api.py (per-process store)	compliance dashboard, alerts, configure-alerts	GET /v1/compliance/dashboard, /alerts, POST /alerts/configure	Compliance view	compliance	Beta
GDPR runtime policy	compliance/gdpr.py	via run-compliance with metadata.purpose	via /v1/compliance/monitor	Compliance view	compliance	Stable
Regulatory watcher	compliance/regulatory_watch.py	— (SDK)	—	—	compliance	Stable
Differential privacy	compliance/advanced.py DPMechanism	— (SDK)	—	—	compliance quickstart	Stable
Framework adapters	integrations/frameworks	— (SDK)	—	—	integrations	Stable
Hallucination detection	evaluation/hallucination.py	— (SDK)	— (MCP check_grounding)	—	cookbook, feature status	Stable
Run tracing	observability/tracing.py	— (SDK)	—	—	tracing	Stable
ModelSession switching	client/model_session.py	— (SDK)	—	—	cookbook	Stable
Agents and orchestrator	agents/	— (SDK)	—	—	architecture	Stable / Beta
Chat	gateway/models.py	chat start	POST /v1/chat, /v1/generate	—	CLI, gateway	Stable
Chat sessions	gateway/chat.py	chat list-sessions, load, save, delete	/v1/sessions, /{id}, /{id}/messages, /{id}/save, /load	—	gateway	Stable
Model compare, health, metrics, list	gateway/	models compare, health, metrics, list	POST /v1/compare, POST /v1/health/check, GET /v1/metrics, GET /v1/models	—	CLI, gateway	Stable
Local model download, export, delete	transformers directly	models download, export, delete	— (CLI only, needs torch)	—	CLI	Beta
Model conversion	model_conversion	convert (needs [finetune])	— (CLI only, needs torch)	—	CLI	Beta
Context transfer	context_transfer	context-transfer	— (CLI only)	—	CLI	Stable
RAG	rag/, gateway/rag_api.py	— (SDK)	/documents, /files, /query, /generate, /models/switch, /token	—	RAG, RAG API	Stable
Memory systems	memory/	— (SDK)	—	—	memory	Stable
Fine-tuning	fine_tuning/	— (SDK)	—	—	features	Beta
MCP server	mcp_server/	python -m multimind.mcp_server	—	—	MCP server	Stable
Backend server	server/platform.py	backend	/, /health, /ready, /docs	—	API overview	Stable

Remaining gaps, by decision

• CLI only, by design: models download, models export, models delete, convert, and context-transfer need torch or local files and are not server features. The CLI page says so.
• SDK only, by design: the guard wrapper, regulatory watcher, differential privacy, framework adapters, hallucination detection, tracing, ModelSession, agents, memory, and fine-tuning are Python APIs. They have docs pages and tests but no CLI or route.
• No dashboard view for chat, sessions, models, and RAG routes. The governance dashboard covers governance artifacts only. A chat playground lives in the separate frontend workspace.
• Compliance history is per process. /v1/compliance/dashboard and /alerts reflect the POST /v1/compliance/monitor calls recorded in that backend process and reset on restart. From the CLI each command is its own process, so compliance dashboard prints nothing recorded. The durable artifacts remain the JSONL audit and cost logs.
• Cross-origin calls from the dashboard page to the backend and the proxy need MULTIMIND_CORS_ORIGINS set on both servers. The views show a curl fallback when blocked.
• models metrics prints empty tables from the CLI because the gateway monitor is per process. Documented on the CLI page.

Fixed on 2026-10-12

• multimind compliance run-compliance no longer exits 1 after printing results.
• multimind --version works.
• Compliance dashboard and alerts record checks instead of rebuilding an empty shard per request.
• compliance run-example names the real missing import.
• Backend gained period, include_env, and the evidence-report filters, plus session save and load routes.
• Dashboard gained scan text, evidence report, compliance, and proxy reload views.
• Guard proxy accepts MULTIMIND_CORS_ORIGINS.
• Docs gained the tracing page, full convert and context-transfer references, and eleven feature-status rows.