Shared context behind human + AI work.
DM keeps what the team decided and how work should be done — Rules, Skills, and Decisions — together with repository intelligence around the task. Memory is on the same path and is not equally live yet.
What matters for this task goes in. What the team learns can come back for review.
What DM brings into the work.
Rules, Skills, Decisions, repository intelligence, architecture context, Prepare Context, and MCP access are ready in a DM workspace. Memory is in progress. Providers are how some capabilities are implemented — they are not the product.
| Capability | Value | Status | Implementation |
|---|---|---|---|
| Decisions | Keep current decisions and rationale available to the work. | Ready | DM |
| Rules | Give humans and connected agents the constraints that apply. | Ready | DM |
| Skills | Reuse proven ways of working. | Ready | DM |
| Repository intelligence | Understand code structure, dependencies, relationships, and architecture around the task. | Ready | Graphify integration |
| Architecture context | Keep architecture boundaries and important system context available around the task. | Ready | Graphify-backed repository intelligence |
| Memory | Carry useful context from previous work across sessions and tasks. | In progress | Mem0 integration |
| Prepare Context | Bring the relevant pieces together for the task at hand. | Ready | DM |
| Agent access | Use DM through supported MCP clients. | Ready | MCP |
| Search / Ask DM | Ask and search the workspace for approved context that applies. | Ready | DM |
DM is not a Mem0 or Graphify wrapper. Graphify is the current repository-intelligence integration. Mem0 is the current memory integration, and that path is still in progress.
Not notes. Governed knowledge.
"Governed" means a person decided this is trusted — not a raw note or an AI guess. Each object below carries who approved it, where it applies, and whether it's still current.
What was chosen and why: rationale, evidence, rejected alternatives, scope, status, lineage.
What must, must not, or needs approval. Versioned, tied to the scope and authority behind it.
How a class of work should be performed: reusable methods, dependencies, validated procedure.
Who may do what, where, and under which conditions — used to inform task context. Runtime enforcement stays with your systems.
Decision → use JSON for machine artifacts
Rule → never store them in Excel
Skill → define schema · validate · store example
Auth → agent may write repo;
external upload needs approval
The entire knowledge base does not belong in every task.
DM resolves the current, in-scope, permitted subset for the work in front of the user or agent. Superseded, irrelevant, or unauthorized knowledge stays out. Conflicts and missing context surface instead of being silently mixed into a prompt.
- Current, approved, in-scope Decisions and Rules
- Skills relevant to the work at hand
- Authorization context where it exists — this object type is still direction, not equally live
- Superseded or expired decisions
- Knowledge outside the task's scope
- Records the actor isn't authorized to see
AI proposes. Humans decide what becomes trusted.
Candidate
Imports, AI-drafted extractions, and agent write-back are all candidates — never approved knowledge on their own.
Review
A person reviews each candidate with its evidence and reasoning in context.
Approve · edit · reject
Approved versions become durable, versioned, and traceable. Nothing else counts as organizational knowledge.
DM is a knowledge & governance product. Your runtime still enforces the actions an agent may take.
The tools a connected agent calls.
Prepare already-approved context that applies to this task. Reads do not wait for a new approval.
Propose new learning for human review. This is a receipt — it does not make the update trusted.
Search approved Decisions in the active workspace.
Submit a Decision candidate for human review.