Early stage, building in public
SALEM is a persistent personal intelligence. It understands your digital world and acts across your tools, devices and services.
AI that lives beyond the chat window.
It speaks first.
Today SALEM runs as a coordinator you reach from a messaging app. It watches the work and services it is connected to, tells you what changed, and keeps your task inbox in order.
The rest of this page is where it is going. This part already works, in the founder's own setup.
Running today Illustrative conversation. Real ones stay private.
- SALEM:
Morning. Two things changed overnight: the backup finished, and the build on main failed after yesterday's last merge. The test job is missing a configuration value.
- SALEM:
I added "Fix the failing build" to your inbox for today.
- You:
Thanks. What else is on today?
- SALEM:
Three items. The build fix comes first. The other two can wait until Thursday. Move them?
- You:
Yes, move them.
- SALEM:
Moved to Thursday. Your inbox for today has one item.
Your AI shouldn't disappear when the chat closes.
Most assistants start from zero every time you open them. SALEM is built around the opposite idea: the people, projects, devices and commitments in your life persist between conversations, so every request starts from what is already known.
It is not another chatbot or another agent framework. Models, agents and tools are parts SALEM uses. SALEM is what stays.
- Your worldProjects, people, devices, services, events, tasks
- SALEMOne persistent context
- OutcomeReasoning and action, wherever you are
An intelligence that understands your world, and shows you what it knows.
Exploring Illustrative example with fictional data.
SALEM keeps a navigable map of your digital world. Ask what it knows about a project, what changed since yesterday, or why it acted. Then correct it, or ask it to forget.
Select any point on the map.
What does SALEM know about Project Atlas?
Open tasks, the decision log and who is involved. Since yesterday: two changes merged, one question waiting on you.
"I'm travelling to Lisbon next week. Make sure everything is ready."
- CalendarTwo overlapping meetings found and flaggedPending
- TravelCheck-in reminder set for MondayPending
- HomeAway mode prepared, waiting for your approvalPending
- InfrastructureRemote access checkedPending
- TasksProject Atlas handover note draftedPending
The right intelligence for every task.
In development Today models are chosen by configuration. Automatic routing is next.
SALEM orchestrates models instead of locking your world to one provider. Models will change. The intelligence layer should not.
- Private or personal data
- Local model
- Stays on hardware you own.
- Complex reasoning and planning
- Claude
- Frontier reasoning when the problem deserves it.
- Cloud and Microsoft workflows
- Azure AI
- Close to the services it works with.
- Writing and changing code
- Coding agent
- A specialist, supervised by SALEM.
Exploring Claude as a reasoning layer. SALEM is evaluating Claude for complex reasoning, coordination between agents, tool use through MCP, and code and infrastructure workflows, while staying multi-model by design.
Understanding is useful. Acting is better.
SALEM sits above models, agents and tools. Any of them can be replaced without SALEM losing what it knows.
What it can reach
Each capability is a layer with its own permissions. Running today means it works in the founder's own setup, not that it is available to others yet.
Messaging
Talk to SALEM from the chat app you already use.
Running todayModels
Local and cloud models, selected per task by configuration.
Running todayTools
MCP servers, APIs and command-line tools, with approvals for anything risky.
In developmentInfrastructure
Watch services, explain failures, propose and apply fixes.
In developmentVoice
Speak to SALEM and hear it answer, at home or on the move.
ExploringDevices and home
Presence across the places and devices around you.
Exploring
Roadmap
What exists, what is being built and what is still an idea. Nothing below is presented as finished.
Now
- SALEM architecture
- Local and cloud foundation
- Agent experiments
- MCP and tool integration
- Multi-model experiments
Next
- SALEM World
- Persistent context
- Model routing
- Voice
- Infrastructure actions
Later
- Proactive intelligence
- Distributed agents
- Telephony
- Presence across devices
- Third-party extensions
Principles
- Persistent
- Context survives individual conversations.
- Provider-agnostic
- No single AI provider owns the intelligence layer.
- Local-first where useful
- Sensitive work can stay on hardware you control.
- Observable
- Model choices and actions can be inspected after the fact.
- User-controlled
- You can see what SALEM knows and what it can reach.
- Extensible
- Tools, agents and providers are modules, not foundations.
About
SALEM is an independent experiment exploring what personal AI could become when intelligence is no longer confined to a chat window.
Why a cat? In folklore, a familiar is a companion that lives with you and helps you. SALEM is meant to be that kind of presence: quiet, attentive, and there when you need it. At rest it slow-blinks, which in cats is a sign of trust.
It is built by Alex Cabrera, an AI engineer focused on agentic systems, automation and AI architecture.