John D. · Private AI Architecture
PRIVATE AI · SECURITY CONTROLS · AWS

OpenClaw Private AI Assistant

Security-focused private assistant reference implementation with session-scoped memory, pluggable model providers, Secrets Manager-backed authentication, allow-listed tools, explicit approval gates, local Docker deployment, and an AWS serverless deployment path.

Amazon BedrockAPI GatewayLambdaDynamoDBSecrets ManagerDockerSQLiteTerraform
View AWS Architecture ↓View Operations Dashboard ↓
PROJECT STATUSImplemented
CI VALIDATIONLoading…
DEPLOYMENTManual by Design
TOOL MODELAllow-listed · Read-only
ASSISTANT TRUST PATH

Authenticate → Orchestrate → Remember → Control → Infer

AUTHAPI Boundary

Bearer token validation protects chat requests.

ORCHESTRATEPolicy-aware Runtime

Validates sessions, messages, and approval requirements.

MEMORYPrivate State

SQLite locally or DynamoDB for AWS deployment.

TOOLSApproved Registry

Read-only guidance and escalation tools only.

MODELBedrock / Local Demo

Bedrock in AWS mode or deterministic offline inference.

AWS ARCHITECTURE DIAGRAM

OpenClaw Private AI Assistant on AWS

Secure serverless request path with API authentication, session-scoped memory, controlled tool access, least-privilege model invocation, and operational logging.

AWS architecture diagram for OpenClaw Private AI Assistant showing API Gateway, Lambda authorizer, Secrets Manager, assistant Lambda, DynamoDB session memory, Amazon Bedrock, allow-listed tools, IAM, CloudWatch, and Terraform
The repository implements this AWS deployment path with Terraform but does not claim the cloud stack is currently running in production. Local Docker mode uses SQLite memory and a deterministic demo provider unless Bedrock is explicitly enabled.
OPENCLAW OPERATIONS DASHBOARD

Private Assistant Security & Runtime Posture

Portfolio dashboard for the implemented control model, deployment posture, memory boundary, tool restrictions, and CI health.

● REFERENCE IMPLEMENTATION
AUTHENTICATIONProtectedBearer-token boundary with Secrets Manager support.
MEMORY MODESession ScopedSQLite locally or DynamoDB in the AWS path.
TOOL POLICYRead-onlyExplicit allow-list; model cannot create new tools.
APPROVAL CONTROLEnforcedHigh-risk or destructive requests stop before execution.

Security control coverage

Enforced
Scoped
Allow-list
Required
Loading…

Control events

Session boundaryConversation state is partitioned by validated session ID.
Secrets boundaryAuthentication token value remains outside source code and Terraform state.
Tool registryOnly approved read-only capabilities are exposed to the assistant.
High-risk requestReturns approval-required guidance instead of automatically executing.
MODEL PROVIDERAmazon Bedrock / Local Demo
AWS DATA STOREAmazon DynamoDB
DEPLOYMENT PATHDocker + Terraform

Implemented controls

Private Session MemoryValidated session IDs partition conversation history.
Least-Privilege IAMOne memory table and one configured Bedrock model.
Secrets BoundaryAPI token value stays out of source and Terraform state.
Tool Allow-listThe model cannot create or authorize new tools.
Approval GatePrivileged or destructive requests are stopped before execution.
Hardened DockerNon-root runtime, dropped capabilities, and loopback publishing.
Higher-risk requests do not execute automatically.The assistant returns an approval-required response and escalation guidance instead.
Reference implementation and CI status only; this page does not claim the Terraform stack is deployed in production.

Recent validation runs

RunResult
Loading…
Reading public GitHub Actions metadata.
MICROSOFT AZURE REFERENCE ARCHITECTURE
Microsoft Azure reference architecture diagram for openclaw private ai assistant

Azure reference equivalent for portfolio comparison. The original project implementation remains as documented elsewhere on this page.