SHAMPOO: introduction

SHAMPOO: introduction

What SHAMPOO is

SHAMPOO (Shared Human-Agent-Model Platform for Orchestration and Operations) is a platform where humans, agents, and models do durable work together. It provides persistent memory, a knowledge graph, tasks and notes, context continuity, hand-offs, reflection, structured debate, quorum workflows, jobs and lanes, human clarification, session continuity, and profile-separated trust boundaries, all reachable over the network from multiple hosts.

SHAMPOO Memory MCP is one product/service surface of that platform: the MCP interface through which local assistants (OpenCode, Codex, Claude Code, and others) read and write shared memory, coordinate tasks, and run structured collaboration. It is not merely a database, a memory server, or a task manager, though it contains all three as parts.

Why it exists

Models forget. Every session starts blank: project context, prior decisions, vendor facts, and half-finished plans evaporate unless someone re-states them. Humans compensate with documents and chat history; agents compensate by re-deriving everything from scratch. Both are slow and lossy.

SHAMPOO exists so that:

• Humans state something once (a decision, a vendor detail, a plan), and every later session, human or agent, can recall it.
• Agents coordinate through durable tasks instead of chat messages that scroll away: create work, claim it, complete it, hand it over.
• Models arriving fresh mid-project can be brought up to speed with a digest, a session recall, and a context pack instead of a week of re-reading.
• Groups can disagree productively (debate), record what was agreed (quorum), and execute it in tracked streams (lanes and jobs), with the reasoning preserved, not just the outcome.

The joined model

A typical SHAMPOO workflow joins every layer:

1. A human asks an agent to investigate something. The agent stores findings as entities and observations (memory) and open questions as tasks.
2. Where the path is unclear, the agent asks for human clarification through the platform instead of guessing.
3. Competing approaches are argued as a structured debate; the decision is recorded as a quorum with named participants.
4. The agreed work becomes lanes and jobs that executors claim and complete, with progress visible to everyone in the boundary.
5. Later, days or months on, reflection passes consolidate the accumulated memory, and a new session recalls all of it in minutes.

No step requires the participants to be online at the same time, to use the same assistant, or to share anything outside their granted profile boundary.

Private versus shared work

SHAMPOO draws one hard line: profile boundaries. A profile is a separate trust and storage boundary served by one endpoint. A principal (client identity) that exists in one profile has no access to another: not read, not write, not even confirmation that content exists there. Failed cross-boundary attempts deny generically, revealing nothing.

Within a boundary, projects scope content further: a principal sees only the projects it was granted. Public collaboration happens in shared projects inside a shared profile; private work stays in private projects or private profiles. See 01-concepts and 03-security-and-boundaries.

Why persistent memory matters

Short-lived context makes agents polite but amnesiac: they re-ask settled questions, contradict recorded decisions, and lose vendor and operational facts between sessions. Persistent memory converts each session from a cold start into a continuation. The cost is discipline (write things down, scope them correctly, retire what is stale), and the platform is built to make that discipline cheap: one tool call to save a session, one to record a decision, one to recall everything later.

Start with 01-concepts for the vocabulary, then 02-architecture for how the pieces connect.