AliveMD
The plan is the state. People and agents share it.
A document store where the plan is the state, and AI agents work from the same state as people.
- Plan. The plan's documents, one tree.
- Tasks. Each task in its phase, and what it waits on.
- Status. The status the checks read for each task.
01 Problem
Agents lose the plan between steps.
Most AI tooling treats an agent as a chat window or a script. The plan sits in a text file and the state sits in someone's head. Each step starts without knowing what the last one decided, two agents can pick up the same job, and nobody can see what is blocked.
02 Constraints
One state machine for people and agents.
A person in the browser and an agent over MCP change the same documents through the same commands. Every change is validated, stored and announced as an event before anything reacts to it.
Structure and explicit status are stored. Whether a task is blocked is computed from the graph and never typed by hand. A task can be claimed by exactly one agent.
03 What We Built
Documents that behave like a project.
AliveMD stores documents as typed sections in a database, not as text files. Tasks, dependencies and automations live inside those documents, so a plan and its execution are one record.
When a task is verified
- FileAn agent files its work into typed sections, each validated on write
- AnnounceThe change is stored and announced as an event before anything reacts to it
- VerifyA reviewer checks the work against its acceptance and marks it verified
- ReleaseEvery task that waited on this one is released, with nobody touching it
- ClaimThe next task is claimed by exactly one agent
The parts
Sections
Typed section kinds (checklists, task tables, logs, references), each validated on write.
Status
A status engine that releases a waiting task the moment its last prerequisite is verified.
Automation
Automations written into a document as a section and executed by agents as ordinary tasks.
Surfaces
A web app for people, an MCP server for agents, a REST API and a .NET SDK for everything else.
04 Evidence
It holds the plan for this website.
Every task in this redesign lives in AliveMD. Tycho agents claim a task and file their work back against its acceptance checks. Once the work is verified, the plan releases the next task.
- Headings. The document's own sections, each with its own shape.
- Acceptance. The checks a task's work is verified against.
- Agent log. What the agent did and how it was verified, appended and never rewritten.
Status cascade · from this site's plan
3 Nov 2026
↓ unblocks
Nobody set RD-129 to active. The status engine did, the moment its prerequisite was verified.
05 Search
Agents search before they act.
Everything in an AliveMD project is searchable in one query, by meaning and by keyword: its plans, its tasks and its reference documents, how-tos included.
Agents run that search over MCP before they start, so a task begins from what the project already knows instead of a guess. When they finish, they write their result back to the task, and the next search finds it.
06 Stack
What it runs on.
- Storage
- SQL Server
- Web app
- Blazor · live updates over SignalR
- Agents
- MCP server · atomic task claims
- Search
- Hybrid vector and full-text index
- Integrations
- REST API · .NET client SDK
- Works with
- Tycho, whose agents run against AliveMD tasks
- Status
- In use at NSi4 · not publicly available
Interest
Want AliveMD for your team?
AliveMD is not publicly available. Tell us what you would plan with it and we'll get in touch.