Keep work moving across steps
Turn an intent into a sequence of decisions, tool calls, checks and handoffs that can continue after one model response.
NAESTRO / CAPABILITIES
The useful unit is the continuing workflow.
NAESTRO brings the pieces around a model into one operating surface: context, planning, tools, governance and evidence. Each capability has a boundary so a workflow can be inspected instead of merely described.
Route the work
Select a capability that fits the next task.
Carry the context
Connect the tools, providers and state it needs.
Recover deliberately
Keep failures visible and the next move explicit.
What the system coordinates
Carry context with the work
Use persistent memory and explicit context boundaries so a later step can see what an earlier step established, changed or left unresolved.
Make consequential work reviewable
Place policy decisions, approvals and refusals alongside the action they govern. A stopped step remains part of the record.
Leave an evidence trail
Review the inputs, tools, policy state and outcome that shaped a trajectory, with the scope of each record stated clearly.
Where it fits
Model
A model generates a response. NAESTRO can place that response inside a longer path with a goal, available tools and a review boundary.
Agent runtime
An agent runtime executes a loop. NAESTRO makes the loop legible across planning, tool activity, context and recovery.
Evaluation protocol
An evaluation defines what a result means. NAESTRO can carry the protocol and its evidence alongside the trajectory.
Supported interfaces, stated precisely
The source adapter layer includes Claude Code, Codex, Gemini, Grok and DeepSeek in optional CLI mode. These are configured process adapters: availability depends on the installed client, explicit enablement, authentication and the adapter contract for that version.
| Mode | Responsibility | Boundary |
|---|---|---|
| CLI adapters | Invoke a configured agent client and retain its result. | Optional; requires client setup and version-specific validation. |
| Local services | Supply working context and tool results. | Configuration and granted access define the available surface. |
| Website example | Explore two fixed trajectory fixtures. | Local browser state; no live agent or external action. |
For the language/runtime layer, see MIND documentation .
For governed persistent memory, see MIND-Mem .
THE LONG VIEW
One operating system. Many kinds of work.
NAESTRO is being shaped as a commercial, tenant-scoped operating environment. Different kinds of work can share a governed trajectory while permissions, data and modules remain scoped. Edge, local, private-fleet and cloud paths can connect through that model; availability depends on each integration.
Work stays in context.
Projects, threads, tasks, tools and results meet in a continuing workspace.
Operations stay governable.
Users, roles, organizations, providers, nodes, modules, usage and audit are the control view.
THE COORDINATION LOOPProduct architecture
Goals & constraintsSet the direction
Models & agentsPropose the work
Memory & knowledgeCarry the context
META BRAINNAESTROPlan · route · coordinate
Admission checksContracts · permissions · required evidence
Tools & servicesConnected software and workflows
Devices & computeEdge, local, fleet and cloud
Results return to shared context.Observe → evaluate → propose improvements → review
ADMITTED EXECUTION TARGETS
EDGEnear the workLOCALoperator-ownedPRIVATE FLEETscoped capacityCLOUDconnected service
FOUNDATION LAYER
The MIND ecosystem keeps the roles distinct.
Contracts, execution, composition and memory are connected through NAESTRO while establishment, authorization, review and evidence keep their own boundaries. The wider MIND direction spans source intent, typed computation and target adapters; support is validated per integration.
MIND-SpeccontractsMIND runtimeexecutionMIND-FlowcompositionMIND-MemcontinuityConcept view. The business OS, tenant boundaries and module integrations proceed through explicit architecture and delivery gates.
A modular business OS
The product direction is one operating environment across leading models, agent CLIs, harnesses and compute nodes. Specialized workspaces bring that foundation to different kinds of work, carrying their own skills, agents, tools and review requirements.
The business-OS architecture is defined; module packaging and delivery proceed in stages. Coding is the first showcase. Research connects to MIND-Lab, while business modules extend the same foundation into customer, document and financial workflows.
First showcase
Coding & engineering
From repository context to implementation, CI, architecture review and a reviewable change.
Integration direction
Research & MIND-Lab
Connect questions, literature, experiments and retained evidence through a continuing research loop.
Module roadmap
Accounting & bookkeeping
Bring documents, reconciliations and period-close preparation into an inspectable workflow.
Module roadmap
CRM & engagements
Carry customer context, follow-ups and engagement history across the life of a relationship.
Module roadmap
Finance & fintech
Coordinate analysis, invoices and financial operations with explicit authority for consequential actions.
Module roadmap
Documents & operations
Connect business knowledge, document work, marketing and audit preparation to the same operating context.
Planned module packaging supports organization-specific workspaces, explicit capabilities and versioned updates. A domain panel, a configured integration and a fully validated business workflow are separate milestones.
Designed to grow across execution systems
Models & CLIs
Use different model families and agent clients according to the task. The adapter contract determines which operations a configured client can perform.
Harnesses & skills
Compose specialized agents, reusable procedures and external execution loops into a reviewable workflow. MIND-Nerve contributes skill routing; MIND-Mem contributes continuing context.
Nodes & compute
The product is designed for edge deployment, with local machines, private fleets and connected cloud resources forming the wider execution environment. Routing must account for available capacity, access and workload requirements.
Workspace & oversight
The frontend is organized around chat, terminals, task views, architecture inspection, memory, evaluation and review. The roadmap extends it with tenant-scoped modules and organization-specific controls.
Start with a real workflow
Bring a workflow with a clear handoff, a consequential decision or a recovery path. We can map the smallest useful trajectory and identify which evidence the first review should contain.
Discuss your workflowBring a real workflow. Start with a bounded question.
Discuss an evaluation