Navin Platform documentation
This is the documentation of the Navin Platform: the portal, the developer guides, MCP, security and governance, the architecture and the bricks. Every page exists in English and in French with the same structure; the French landing page is ../fr/README.md. The full map of the site is ../index.md.
Start here
- Getting Started: a first secure agent
- Portal user guide
- Developer guides
- API reference: Control Plane API (
/api/control-planeon the documentation site) and Navin Manager API (/api/manager)
Portal user guide
Overview
- Overview (
/)
Administration
- Tenants (
/admin/tenants) - Projects and environments (
/admin/projects) - Organizations and business units (
/admin/organizations) - Members and roles (
/admin/members) - API keys (
/admin/api-keys) - Service identities (
/admin/service-identities) - Directory groups (
/admin/group-mappings) - Licence and support (
/admin/license)
LLM management
- Providers (
/llm/providers) - Models registry (
/llm/models) - Router and fallbacks (
/llm/router) - LiteLLM config (
/llm/litellm) - LLM projects (
/llm/projects) - LiteLLM keys (
/llm/keys) - Model versions (
/llm/versions) - Isolation mode (
/llm/isolation) - Capability resolver (
/llm/resolver) - Context budget (
/llm/context) - Gateway playground (
/llm/playground)
Agents
- Registry (
/agents/registry) - Runs (
/agents/runs) - Approvals inbox (
/agents/approvals) - Schedules (
/agents/schedules) - Triggers (
/agents/triggers) - Sessions (
/agents/sessions) - Applications (
/agents/applications)
Capabilities
- Tools (
/capabilities/tools) - MCP servers (
/capabilities/mcp) - Skills (
/capabilities/skills) - Plugins (
/capabilities/plugins) - Connectors (
/capabilities/connectors) - Channels (
/capabilities/channels) - Webhook deliveries (
/capabilities/webhooks)
Marketplace
- Pack catalog (
/marketplace/catalog) - Installations (
/marketplace/installs) - Change pipeline (
/marketplace/changes)
Data and knowledge
- Knowledge bases (
/data/knowledge) - Memory (
/data/memory) - DLP tester (
/data/dlp) - Retention (
/data/retention) - Lineage (
/data/lineage) - Prompts (
/data/prompts) - RAG quarantine (
/data/quarantine) - Storage profiles (
/data/storage)
Multimodal
- Media library (
/multimodal/media) - Jobs (
/multimodal/jobs) - Voice live (
/multimodal/voice) - Voice consents (
/multimodal/consents)
Quality
- Evaluation suites (
/quality/suites) - Evaluation runs (
/quality/runs) - Quality gates (
/quality/gates) - Red team (
/quality/redteam) - A/B comparisons (
/quality/comparisons)
Security and governance
- Security posture (
/security/posture) - Policies (
/security/policies) - Decisions log (
/security/decisions) - Governance registry (
/security/governance) - Risk and compliance (
/security/risk) - Incidents (
/security/incidents) - Secrets (
/security/secrets) - Encryption keys (
/security/encryption-keys) - SOC and AI-SDR (
/security/soc) - Kill Switch (
/security/kill-switch) - Delegations (
/security/delegations) - Permissions (ReBAC) (
/security/permissions) - Break glass (
/security/break-glass) - Policy simulator (
/security/policy-simulator) - File security (
/security/file-security)
Observability
- SLO and error budgets (
/observability/slo) - Live metrics (
/observability/metrics) - Traces (
/observability/traces) - Request journal (
/observability/logs)
FinOps
- Summary (
/finops/summary) - Usage events (
/finops/usage) - Price cards (
/finops/price-cards) - Quotas (
/finops/quotas) - GPU allocation (
/finops/gpu) - Reconciliation (
/finops/reconciliation) - Alerts (
/finops/alerts) - FinOps settings (
/finops/settings)
Audit
- Events and chain (
/audit/events) - Correlations (
/audit/correlations) - Evidence and seals (
/audit/evidence) - Legal holds (
/audit/legal-holds)
Developer portal
- Developer kit (
/developer/kit) - Contracts (
/developer/contracts) - API reference (
/developer/api) - SDK quickstarts (
/developer/sdk) - Brick catalog (
/developer/bricks) - Roadmap (
/developer/roadmap)
Developer Studio
- Prompt studio (
/studio/prompts) - Skill studio (
/studio/skills) - MCP studio (
/studio/mcp) - Connector studio (
/studio/connectors) - Plugin builder (
/studio/plugins) - Policy studio (
/studio/policies) - Navin library (
/studio/navin-library)
Developer guides
- Getting Started: a first secure agent
- Agent Development Guide
- MCP Development Guide
- Hosting MCP servers
- MCP Runtime per user and project
- MCP security
- Skill Development Guide
- Plugin Development Guide
- Pack development guide: plugins, MCP servers, skills, prompts and connectors
- Connector Development Guide
- Prompt Development Guide
- Policy Development Guide
- Testing Guide
- Migration Guide
- Marketplace of complete packs
- Navin library: skills, plugins, MCP, channels and rules per project
- File security: ClamAV and YARA-X before any ingestion
- Client storage administration (brick E-13)
- Developer portal and living documentation
- API Reference (OpenAPI and examples)
- ExecutionContext Reference
- ActionContext Reference
Security and governance
- Enterprise AI Platform security
- Approvals and human oversight
- Audit Gateway and Evidence Service: security model
- Authentication and authorization
- Security configuration
- RAG data protection
- File security pipeline: security model
- Governance of the Navin Platform
- Harden a production installation
- Identity inherited end to end: Context Authority, delegations, mandates, break glass
- Identity and policy plane: operator guide
- Injection and attack matrix
- Retention, privacy and personal data protection
- Risk, compliance and evidence
- Security incident runbooks
- Secrets, encryption keys and IT vault
- SOC, AI-SDR, Kill Switch and recovery runbook
- Security testing
- Threat model
Architecture
- Enterprise AI Platform - Architecture overview
- Platform architecture
- PostgreSQL data model
- Target monorepo mapped to the real repository
Architecture decision records
- ADR-0101: Licence policy of the distribution
- ADR-0102: Navin runtime without a declared graph
- ADR-0103: vLLM and LiteLLM roles, LiteLLM pinned to 1.106.5
- ADR-0104: OpenAI compatible default call format, calls through the Model Gateway
- ADR-0105: Reasoning effort never converted silently
- ADR-0106: Platform, tenant, project, environment tenancy with forced RLS
- ADR-0107: FinOps token rule and three amounts
- ADR-0108: Opik instead of Langfuse
- ADR-0109: Perses instead of Grafana, superseded by ADR-0169
- ADR-0110: OpenSearch instead of Loki
- ADR-0111: SeaweedFS and client storage instead of MinIO
- ADR-0112: OpenBao or client vault (MPL-2.0 exception)
- ADR-0113: Redis as cache only
- ADR-0114: PostgreSQL as source of truth
- ADR-0115: Context Authority and delegation intersection rule
- ADR-0116: Audit hash chain, seals and outbox
- ADR-0117: Kill Switch registry, desired and observed state on receipts
- ADR-0118: Signed licences without price or limit
- ADR-0119: Go node agent and Kubernetes operator tested with envtest
- ADR-0120: Marketplace of signed packs
- ADR-0121: Egress guard as the only outbound path
- ADR-0122: English and French everywhere
- ADR-0123: Test honesty: PASSED only with evidence
- ADR-0124: SSH only for bootstrap, operations through agents, operator and APIs
- ADR-0125: Service ownership (PLATFORM_MANAGED, CLIENT_MANAGED) and six deployment modes
- ADR-0126: Configured high availability is not validated high availability
- ADR-0127: Backup, restore and DR separate from HA, on the client's storage
- ADR-0128: Single console with a Platform Manager space, standalone Manager console kept
- ADR-0148: Portal frontend stack (design system, charts, graphs, forms, API client)
- ADR-0149: Server routes of the B9.3 SDK interfaces (service identity, application audit, approval cancel, storage objects)
- ADR-0150: Internal operation engine of the Platform Manager instead of Temporal
- ADR-0151: Compatibility matrix held by the signed catalog, not by a compatibility_matrix table
- ADR-0154: Audit index partitioned by event_time and view under the reference names
- ADR-0155: Worker planes, operation_id of long operations and compensation of external actions
- ADR-0156: Tenant hierarchy, common fields stamped by PostgreSQL and a separate runtime login
- ADR-0157: Bounded span summary in PostgreSQL and SLO objectives in the database
- ADR-0158: SLSA provenance, G6/G9 gates, pack contents, kill switch and offline import of the marketplace
- ADR-0159: Single change management pipeline
- ADR-0162: Backup and DR, configuration center and deep links in the portal Platform Manager space
- ADR-0163: One dedicated test per portal screen and a manifest based plugin builder
- ADR-0164: Navin library activated per project, project rules, MCP servers and channels attached to a project
- ADR-0165: MCP servers hosted in the Navin Manager and isolation per project of the MCP gateway
- ADR-0166: High availability declared per service in the Navin Manager catalog, mandatory ClamAV and YARA-X file scanning before any ingestion
- ADR-0167: MCP Runtime isolated per tenant, project and user on the ToolHive Kubernetes Operator
- ADR-0168: Two managed local inference engines, vLLM and Ollama, behind LiteLLM
- ADR-0169: Perses and Infisical OSS removed: dashboards in the Navin Manager and the portal, secrets in OpenBao or the client vault
- ADR-0170: Internal MCP servers only, company MCP catalogue in SeaweedFS, destroyed on failure, woken on call and traceability by default
- ADR-0171: Hardened installation for banking production: isolated network, Quadlet, security profiles, Trusted Hub, rollback
- ADR-0172: Data labeling and model fine-tuning: Label Studio and LLaMA-Factory in the catalog
Implementation status and evidence
- Architecture
- Conformance to the product architecture
- Blockers
- Managed components catalogue (ManagedComponent)
- Decisions
- Project handover
- Dependency licence report (SPDX SBOMs)
- Release readiness
- Platform requirements
- Lot roadmap
- Platform status
- Test evidence
- Audit files of the kit verification lots
- Conformance lot files
Bricks
- Platform bricks
- A01 - Central Portal
- A02 - Control Plane API
- A03 - Human Identity
- A04 - Machine Identity
- A05 - Authorization
- A06 - Policy Engine
- A07 - Policy Enforcement
- A08 - Secrets and Transit Crypto
- A09 - Certificates
- A10 - Multi-tenant Management
- A11 - Governance Registry
- A12 - Approval Workflow
- A13 - Risk and Compliance
- A14 - Incident Management
- B01 - Unified LLM Proxy
- B02 - Local Inference
- B03 - Provider Registry
- B04 - Model Registry
- B05 - Model Router
- Model modalities matrix (B-06)
- B06 - Model Capability Resolver
- B07 - Reasoning/Effort Policy
- B08 - Context Window Manager
- B09 - Token Accounting
- B10 - Model Pricing Registry
- B11 - Cost Reconciliation
- B12 - Model Evaluation
- B13 - Local GPU Management
- B14 - Provider Failover
- C-01 Agent Runtime
- C-02 Durable Execution
- C-03 Agent Loop Controller
- C-04 Sub-Agent Manager
- C-05 Agent-to-Agent
- C-06 Checkpoint Manager
- C-07 Session Manager
- C-08 Heartbeat Monitor
- C-09 Scheduler
- C-10 Event Triggers
- C-11 Human-in-the-Loop
- C-12 Execution Sandbox
- C-13 Resource Limits
- C-14 Agent Kill Switch
- C-15 Execution Replay
- D-01 Tool Registry
- D-02 Tool Gateway
- D-03 MCP Registry
- D-04 MCP Gateway
- D-05 Skill Registry
- D-06 Skill Loader
- D-07 Plugin Registry
- D-08 Plugin Runtime
- D-09 Connector Registry
- D-10 Connector Hub
- D-11 Credential Broker
- D-12 Webhook Gateway
- D-13 Channel Gateway
- D-14 API Gateway
- E-01 Data Classification
- E-02 DLP Gateway
- E-03 Tokenization Service
- E-04 Encryption Service
- E-05 Document Processing
- E-06 RAG Pipelines
- E-07 Vector Database
- E-08 Relational Storage
- E-09 Context Manager
- E-10 Memory Manager
- E-11 Data Lineage
- E-12 Data Retention
- E-13 Object Storage
- F-01 Vision
- F-02 Image generation and editing
- F-03 Video processing
- F-04 Video generation
- F-05 Speech to text
- F-06 Speaker diarization
- F-07 Text to speech
- F-08 Voice live (duplex)
- F-09 Voice activity detection
- F-10 Audio streaming
- F-11 Multimodal DLP
- F-12 Media storage and processing
- G-01 Distributed tracing
- G-02 Telemetry pipeline
- G-03 AI tracing (prompts, responses, tools, scores)
- G-04 Analytics store
- G-05 Metrics
- G-06 Logs
- G-07 Dashboards
- G-08 Alerting
- G-09 Evaluation
- G-10 Red teaming
- G-11 Audit evidence (WORM)
- G-12 Security monitoring
- G-13 Quality orchestrator
- G-14 Audit correlation
- H-01 Quota service
- H-02 Cost ledger
- H-03 Pricing engine
- H-04 GPU accounting
- H-05 Autoscaling
- H-06 Kubernetes platform
- H-07 Networking
- H-08 GitOps
- H-09 Infrastructure as code
- H-10 Policy admission
- H-11 Image scanning
- H-12 SBOM
- H-13 Signing
- H-14 Backup and disaster recovery
- H-15 CI/CD
- H-16 Artifact registry
- Governed applications and third party agents
Installation
- Install the Navin Platform on an air-gapped site
- Run the Navin Platform with Docker Compose
- Configure the Navin Platform for the first time
- Hosting topologies
- Install the Navin Platform
- Install the Navin Platform on Kubernetes with Helm
- Run the Navin Platform for local development with make
- Network ports and flows
- Install the Navin Platform from OS packages
- Sizing and hardware
- Configure the TLS certificates
- Uninstall the Navin Platform
- Upgrade the Navin Platform
- Verify the installation
Reference
SDKs, API, operations and releases
- Enterprise AI Platform SDKs
- Control Plane API v1 - contract shared by backend, portal, SDKs, CLI
- Contrats API principaux de l'architecture / Main API contracts of the architecture
- Operations - Enterprise AI Platform (English)
- Enterprise AI 1.0.0 release notes