Memory

Memory persists context across tasks, loop runs, and chat sessions. Polpo has two scopes:

Scope	Visible to	Local file
Shared	Every agent in the project	`.polpo/memory.md`
Agent	Only the named agent	`.polpo/memory/{agent-name}.md`

Polpo Cloud stores the same logical scopes in the project state store. At the start of each run or chat request, the runtime adds non-empty memory to the system context as <shared-memory> and <agent-memory> blocks.

Agent tools

Enable memory_* when an agent should update its private memory:

Tool	Operation
`memory_get`	Read private memory
`memory_save`	Replace private memory
`memory_append`	Append a timestamped note
`memory_update`	Find and replace a section

{
  "name": "backend-dev",
  "allowedTools": ["read", "write", "edit", "bash", "memory_*"]
}

These tools are scoped to the calling agent. They cannot read or change another agent’s private memory.

API

Project API keys can manage both scopes:

Method	Path	Description
`GET`	`/v1/memory`	Read shared memory
`PUT`	`/v1/memory`	Replace shared memory
`GET`	`/v1/memory/agent/{agentName}`	Read one agent’s memory
`PUT`	`/v1/memory/agent/{agentName}`	Replace one agent’s memory

Write operations accept { "content": "..." }. Keep memory short and structured because its complete content consumes context on every run.

Getting Started

Platform

Agents

Orchestration

Agent tools

API

​Agent tools

​API

Agent tools

API