spectron-strands-agents

Spectron agent memory as tools for the Strands Agents SDK.

Spectron is SurrealDB's provenance-first, tri-temporal memory and knowledge layer for AI agents. This package exposes Spectron's memory operations as Strands tools, so any Strands agent can store and retrieve long-term memory with a single line of setup:

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-agents

pip install "spectron-strands-agents[bedrock]"

Requires Python 3.10 or newer.

Configuration

The tools need a Spectron client. Provide the connection details as environment variables:

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

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

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

Spectron 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 scope for all tools, and let the agent override it per call when needed:

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

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

How it works

Each tool is a thin wrapper over one method on the synchronous surrealdb.Spectron client. Because that client is synchronous, the tools call it directly. There is no background event loop or async shim to manage.

Spectron is in early preview. The remember, recall and context tools follow the documented client signatures. The reflect, forget, upload and inspect tools cover operations whose keyword arguments are still settling. If a method name or argument differs in the version of surrealdb you have installed, the fix is a single call inside the matching factory in src/spectron_strands/tools.py.

Examples

Runnable examples live in examples/. They use Amazon Bedrock, which is Strands' default model provider. See examples/README.md for what each one shows and how to run it.

Roadmap

• Async tools built on AsyncSpectron.
• Streaming chat and the grouped resources (sessions, entities, and others).

License

Apache License 2.0. See LICENSE.