DecentraMind Labs

Overview

DecentraMind Labs is building DecentraMind OS — an AI-native platform focused on the software infrastructure surrounding intelligent agents, rather than treating a model call as the application.

The current stack combines a Next.js application, Express backend, PostgreSQL/Prisma persistence, workspace and project workflows, a deterministic coordination layer, agent identity and Agent OS execution primitives, policy-gated tools, execution/audit records, and optional Solana verification paths.

Active engineering platform — not a completed autonomous-agent product.

Engineering Problem

Calling an LLM is not enough to create an autonomous software system. A usable agent platform also needs identity, authentication, state, workflow context, permissions, tool boundaries, execution state, error handling, scheduling, auditability, coordination, integrations, and — where useful — optional verification.

The challenge is assembling those layers so probabilistic model output can participate in real workflows without becoming unrestricted system authority.

My Role

Founder & Blockchain / AI Architect

I work across architecture and hands-on implementation: application/backend design, PostgreSQL/Prisma persistence, agent-runtime architecture, tool-policy systems, wallet verification flows, coordination infrastructure, implementation planning, code review, and product development.

Contributions are labeled explicitly below. This is not a claim that every code path was solely authored by me.

System Architecture

DecentraMind OS is layered so application workflows, coordination, agent execution, persistent state, and optional verification remain separable concerns.

Model vs Agent Architecture

Inference is one capability among several — not the definition of an agent.

• Model — inference / reasoning
• Agent Runtime — execution coordination
• Memory / State — persistent context outside the model
• Tools — controlled external capabilities
• Permissions — deterministic execution boundaries
• Execution Records — auditability

A provider boundary with optional OpenAI integration exists; real provider use is environment-gated. Dry-run / placeholder paths remain part of active development.

Roadmap / Product Direction Local / self-hosted model providers are product direction — not an implemented provider in the current runtime.

Authentication + Wallet Verification

Primary application authentication is JWT / session account authentication through the Express backend.

Separately, Solana challenge–response wallet linking verifies wallet ownership: an authenticated user requests a challenge, the wallet signs, the backend verifies with Ed25519, and the wallet is linked to that account.

• Solana challenge–response verification for wallet linking Implemented
• Solana wallet-login authentication as primary login Not Production-Complete

Application login and wallet ownership verification are separate concerns. Production Solana wallet-login remains incomplete.

The runtime turns probabilistic model suggestions into observable software execution behind deterministic boundaries. Persisted primitives include agent runs, ordered steps, and tool invocations with allow/deny outcomes, correlated to execution/audit logging.

• AgentOSRun · AgentOSStep · AgentOSToolInvocation records Implemented
• Bounded execution loop, timeouts, dry-run / gated LLM behavior Active Development

Not presented as production-complete autonomous execution.

Probabilistic model output should not equal unrestricted system authority. The runtime exposes approved tools through a whitelist registry, Zod contracts, and deny-by-default policy checks — with allow/deny outcomes recorded on each invocation.

Agent memory structures are persisted alongside agent profile metadata in PostgreSQL/Prisma, separating model inference from application-owned agent state. Not a mature standalone memory product; “user-owned memory” is not claimed as completed.

Runtime heartbeat/state infrastructure exists. Task-queue and scheduling infrastructure is early-stage — not a production scheduler fleet.

AI execution remains off-chain. When configured, an execution digest can be optionally anchored on Solana devnet as a verification/proof layer — not as the compute substrate.

Not mainnet production verification; not blanket on-chain verification of every action.

Domain-specific applications explored on DecentraMind OS include Care Orchestrator, Autonomous CFO, and Crypto Alpha Assistant — application surfaces and CFO simulation/runtime workflows, not production autonomous agents.

They demonstrate shared runtime, state, tools, permissions, coordination, and execution records across different domains.

• Care Orchestrator
• Autonomous CFO
• Crypto Alpha Assistant

Engineering Decisions

PostgreSQL + Prisma

Persistent relational state across users, workspaces, projects, agents, runs, steps, tool invocations, and platform records.

Next.js + Express separation

Frontend/product surface separated from backend platform and runtime services.

Deterministic coordination first

Structured coordination signals operate without LLM generation, providing a stable software foundation before agent-generated enrichment.

Policy-gated tools

Model output is separated from privileged execution through whitelisted capabilities and explicit policy checks.

Model provider boundary

The runtime separates model access from agent state and execution, reducing coupling to a single provider and leaving room for future model options. Optional OpenAI integration is environment-gated; local/self-hosted providers remain roadmap.

Off-chain AI + selective blockchain verification

AI workloads remain off-chain; blockchain is used only where proof/verification adds value.

Solana devnet

Optional low-cost development path for digest anchoring — not a claim of mainnet production verification.

Security & Reliability

Core principle: probabilistic model output should not equal unrestricted system authority.

• JWT / session authentication
• Workspace access checks
• Solana challenge–response wallet linking
• Deny-by-default tool policy · whitelist registry · Zod contracts
• Allow/deny outcomes and execution records persisted
• Coordination activity / audit logs
• Bounded runtime concepts (timeouts, cancellation, dry-run paths)

Not a claim of complete production IAM or mature production agent isolation.

What I Personally Built

• Application architecture for DecentraMind OS Architecture / Implementation
• Next.js + Express platform structure Implementation / Architecture
• PostgreSQL / Prisma persistence model Implementation
• Coordination layer infrastructure Implementation / Architecture
• Agent OS runtime model Architecture / Active Development
• Runs / steps / tool-invocation model Implementation / Architecture
• Policy-gated tool registry Implementation / Contribution
• Wallet-link verification architecture Implementation / Architecture
• Solana devnet proof anchoring architecture Active Development / Architecture
• Technical planning, implementation review, and product development Leadership

Labels reflect repository-grounded contributions. Sole authorship is not claimed for every path.

Technology

• Next.js
• React
• TypeScript
• Node.js
• Express
• PostgreSQL
• Prisma
• JWT / Session Auth
• Zod
• Solana
• OpenAI Integration
• Jest

Verified in the current DecentraMind OS implementation. EIP-191, Solidity, and Ethers.js are not listed as DecentraMind core-platform claims here.

Current Status

Active Development · 2025 – Present