Memory API¶

Persistent agent memory is the pydantic-ai-harness Memory capability over a FileStore in the run's workspace. Enable it via include_memory=True (default) on create_deep_agent, which builds it with build_memory_capability. See Memory for the conceptual overview.

pydantic-deep's own memory API - AgentMemoryToolset, MemoryCapability, MemoryFile, load_memory, format_memory_prompt - is deprecated: it still imports, with a DeprecationWarning, and nothing uses it.

build_memory_capability¶

`pydantic_deep.features.memory.store.build_memory_capability(*, agent_name, store=None, memory_dir=DEFAULT_MEMORY_DIR, namespace='', max_lines=None, max_tokens=None, max_memory_size=None, pin_marker=None)` ¶

The Memory capability one agent remembers with.

Shared by the main agent and every subagent, so their wiring cannot drift.

Parameters:

Name	Type	Description	Default
`agent_name`	`str`	The agent's own segment in the store; sanitized for the harness.	required
`store`	`FileStore \| None`	The store to remember in. Agents writing to one directory should share one: its copies share a lock, which keeps the receipts beside the notebooks consistent. A new `FileStore(memory_dir)` when `None`.	`None`
`memory_dir`	`str`	Directory in the run's workspace, when `store` is not given.	`DEFAULT_MEMORY_DIR`
`namespace`	`MemoryNamespace`	A per-tenant segment above the agent's, fixed or resolved per run. The model never sees or chooses it.	`''`
`max_lines`	`int \| None`	Most `MEMORY.md` lines injected; the harness default when `None`.	`None`
`max_tokens`	`int \| None`	Approximate ceiling on the injected section; likewise.	`None`
`max_memory_size`	`int \| None`	Largest memory file, in characters, the tools will read or change; the harness default (65,536) when `None`.	`None`
`pin_marker`	`str \| None`	Accepted only to warn - the harness has no pinned section.	`None`

Raises:

Type	Description
`ValueError`	A fixed `namespace` the harness would refuse.

MemoryNamespace¶

`pydantic_deep.features.memory.store.MemoryNamespace = str | Callable[[RunContext[Any]], str]` `module-attribute` ¶

A fixed namespace, or one resolved from each run's context - a user id, say.

get_memory_path¶

`pydantic_deep.features.memory.service.get_memory_path(memory_dir, agent_name)` ¶

Compute the memory file path for an agent.

Parameters:

Name	Type	Description	Default
`memory_dir`	`str`	Base directory for memory files.	required
`agent_name`	`str`	Agent name (e.g., "main", "code-reviewer").	required

Returns:

Type	Description
`str`	Full path like ".deep/memory/main/MEMORY.md".