self.recall(): CrewAI method reference

What is self.recall()?

self.recall(scope=None, limit=10) queries CrewAI's unified memory layer — typically backed by a vector database plus metadata — and returns the strongest matches for the current semantic context. Scopes are hierarchical path prefixes such as /agent/researcher or /tenant/acme/support so you can isolate customer-specific facts from global playbooks. The scorer blends embedding similarity with recency and importance weights so stale bulletins decay naturally without manual deletes.

Use recall at the beginning of a flow step or agent turn when you need institutional memory: prior decisions, user preferences, or summarized artifacts produced by earlier runs. Keep payloads small: retrieve snippets and pointers, not entire PDFs, unless your downstream task truly needs them. When combined with self.remember(), treat the pair like a transactional read-after-write — write summaries with explicit scopes so future recall queries stay precise.

Recall is not a substitute for structured Flow state that must be deterministic within a single kickoff. State belongs in the Pydantic model attached to the flow; memory belongs to knowledge that should span runs or cross agents. Blurring the two creates heisenbugs where a recalled paragraph contradicts freshly computed state because embeddings matched incidental wording.

Use Cases

• Cross-run context
• Personalization
• Replaying decisions

Key Features

✓ Composite scoring
✓ Scope-narrowing
✓ Configurable depth

When NOT to Use

When the data lives in the Flow state — read state directly.

Notes

Scope too broad

Omitting scope searches the entire corpus and surfaces unrelated high-similarity noise. Start narrow, widen only when metrics show recall@k is too low.

Embedding drift

If you change embedders or dimensions, reindex before trusting scores. Mixed embedding spaces produce false positives that look like semantic matches.

Latency budgets

Each recall is a vector query. For real-time UIs, prefetch memories in parallel with tool calls or cache hot scopes in Redis keyed by tenant.

Privacy

Memories may contain PII from prior user sessions. Apply the same retention and access controls as your primary database; recall bypasses application-level ORMs.

Parameters

Parameter	Type	Required	Purpose
scope	str \| None	No	Sub-tree to search.
limit	int	No	Max results to return.

Code Examples

python

items = self.recall('/project/alpha', limit=5)

Prefetch into flow state

python

@listen(start_step)
def hydrate(self):
    snippets = self.recall(scope='/playbook/support', limit=4)
    self.state.memory_hits = snippets
    return self.state

When to Use

When an agent needs context from prior runs.

Common Mistakes

❌ Forgetting to set scope and getting noise

✅ Always scope queries to the relevant sub-tree.

self.recall() FAQ

What is self.recall() in CrewAI?

Retrieves stored memories ranked by semantic similarity, recency, and importance. self.recall(scope=None, limit=10) queries CrewAI's unified memory layer — typically backed by a vector database plus metadata — and returns the strongest matches for the current semantic context. Scopes are hierarchical path prefixes such as /agent/researcher or /tenant/acme/support so you can isolate customer-specific facts from global playbooks. The scorer blends embedding similarity with recency and importance weights so stale bulletins decay naturally without manual delet…

Which CrewAI types expose the method self.recall()?

DevShelfHub documents self.recall() on Flow, Agent. The reference maps it to Python module crewai.flow.flow.Flow — pin your installed crewai version and match imports to the snippet on this page.

When should I use self.recall()?