Strands Agents | Spectron

Spectron exposes its memory operations as Strands Agents tools, so any Strands agent can store and retrieve long-term memory with a single line of setup.

Package: spectron-strands (PyPI). It pulls in strands-agents and surrealdb (which provides the Spectron client).

from strands import Agent
from spectron_strands import spectron_tools

agent = Agent(tools=spectron_tools())

agent("Remember that we signed a contract with Meditech Solutions for 1.2M GBP.")
print(agent("What is the value of the Meditech Solutions contract?"))

Installation

pip install spectron-strands

To run the examples with Amazon Bedrock (Strands' default model provider):

pip install "spectron-strands[bedrock]"

Requires Python 3.10+.

Environment

With these set, spectron_tools() builds the client for you:

export SPECTRON_ENDPOINT="https://api.spectron.example"
export SPECTRON_API_KEY="your-bearer-token"
export SPECTRON_CONTEXT="acme-prod"

You can also pass the values directly, or hand in a client you already have:

from surrealdb import Spectron
from spectron_strands import spectron_tools

# From explicit arguments.
tools = spectron_tools(
    endpoint="https://api.spectron.example",
    api_key="your-bearer-token",
    context="acme-prod",
)

# Or reuse an existing client.
client = Spectron(context="acme-prod", endpoint="...", api_key="...")
tools = spectron_tools(client=client)

Tools

spectron_tools() returns seven tools by default, one per Spectron operation:

Tool	Purpose
`spectron_remember`	Store a fact or observation for later recall.
`spectron_recall`	Search memory and return the most relevant stored information.
`spectron_context`	Assemble a working-memory context block for a query.
`spectron_reflect`	Run a synthesis pass that consolidates and connects memories.
`spectron_forget`	Delete memories that match a query.
`spectron_upload`	Ingest a document so its contents become recallable.
`spectron_inspect`	Browse the substrate as queryable data for debugging or audit.

Choose a subset with include or exclude:

# Only the everyday read/write pair.
tools = spectron_tools(include=["remember", "recall"])

# Everything except deletion and inspection.
tools = spectron_tools(exclude=["forget", "inspect"])

Scopes

Scopes isolate memory by principal, tenant, or session. A scope is a path string such as "org/acme/user/alice", or a list of paths. Set a default for all tools and let the agent override it per call:

tools = spectron_tools(scope="org/acme/user/alice")

Every tool also accepts a scope argument, so a multi-user agent can direct a single operation at a specific principal.

When to use MCP or the SDK instead

For an MCP-native host, use the MCP server.
To call Spectron directly, use the Python SDK.