Argentum — on-chain trust layer for ADK agents (karma + community attestations)

[giskard09]

The problem

When ADK agents call external services autonomously, how do you know if a service is trustworthy? Not "does it have permission" — but "has it proven it acts well over time?"

What Argentum does

Every tool call through ADK can now:

• Check service karma on Arbitrum before calling — blocks services with negative karma
• Submit a verified action to build the calling agent's on-chain reputation

The flywheel: every tool call becomes a trace. Traces accumulate into karma. Karma unlocks trust.

One callback, all your tools

from google.adk.tools import FunctionTool
from argentum.adk_plugin import make_trust_callback

trust = make_trust_callback("my-agent-001", "My Agent")

agent = Agent(
    model="gemini-2.5-flash",
    tools=[
        FunctionTool(fetch_price, before_tool_callback=trust),
        FunctionTool(search_repos, before_tool_callback=trust),
    ]
)
# Every tool call now checks karma + builds reputation automatically

Try it (no Gemini key needed)

pip install git+https://github.com/giskard09/argentum-sdk requests
python examples/adk_trust_demo.py

Standalone demo runs against the live API and shows the plugin in action in ~10 seconds.

Stack

• Karma economy running on Arbitrum One (contract 0xD467CD1e34515d58F98f8Eb66C0892643ec86AD3)
• Actions verified by community attestations (weight-based, not binary)
• Works with any ADK agent — no changes to your agent logic
• Apache 2.0

Repo: https://github.com/giskard09/argentum-sdk
Live API: https://argentum.rgiskard.xyz
Dashboard: https://giskard09.github.io/argentum-web/

Open to feedback — especially on the karma thresholds and what other trust signals would be useful for ADK agents.