Framework integrations (Strands & CrewAI)¶

Beyond LangGraph and LiteLLM, memory-reuse ships cached_tool decorators for the Strands Agents and CrewAI frameworks. Both mirror the LangGraph cached_tool surface exactly — same signature, same keying, same scope resolution, TTL expiry, and exact-vs-semantic routing. Under the hood each delegates to the shared caching machinery, so behaviour is identical across frameworks by construction; no cache logic is re-implemented.

Both decorators share this signature:

cached_tool(cache, *, scope="global", ttl=300, semantic=False, exact_only=False)

Parameter	Default	Meaning
`scope`	`"global"`	Scope isolation — `"global"`, `"user"`, or `"session"`.
`ttl`	`300`	Time-to-live in seconds for a stored result.
`semantic`	`False`	Route through the combined exact-then-semantic flow.
`exact_only`	`False`	Force exact-only matching even when semantic is enabled.

A decorated tool returns the stored result without executing its body on a cache hit, runs and stores the result on a miss, and re-executes once the TTL elapses. If the tool body raises, the exception propagates and nothing is stored. A scope that requires a scope ID which cannot be resolved raises ScopeViolationError before the body runs. See Backends & scopes for scope details.

Strands¶

Requires the strands extra:

pip install "memory-reuse[strands]"

from memory_reuse import MemoryCache
from memory_reuse.integrations.strands import cached_tool

cache = MemoryCache()

@cached_tool(cache, scope="global", ttl=600)
async def fetch_weather(city: str) -> dict:
    return await weather_api.get(city)

CrewAI¶

Requires the crewai extra:

pip install "memory-reuse[crewai]"

from memory_reuse import MemoryCache
from memory_reuse.integrations.crewai import cached_tool

cache = MemoryCache()

@cached_tool(cache, scope="session", ttl=300)
async def fetch_user_profile(user_id: str) -> dict:
    return await db.get_user(user_id)

CrewAI adds one decoration-time guard: exact_only=True and semantic=True are contradictory, so supplying both raises ConfigurationError rather than silently ignoring one argument.

Missing-dependency errors

Using either integration without its dependency installed raises BackendNotAvailableError whose message names the extra to install (pip install "memory-reuse[strands]" or pip install "memory-reuse[crewai]") rather than surfacing a raw ImportError. The core package imports fine without either installed.

Semantic matching

Set semantic=True (with semantic_enabled=True on the cache config) to let reworded-but-equivalent calls reuse a stored result. See the Semantic cache guide.