Tide Runtime SDK

Python SDK for wrapping production agent tools with Tide policy, approvals, and earned-use logging.

Recommended from an agent repo:

npx -y --package rippletide-package@latest tide run

That command creates a local .venv, installs tide-runtime inside it, scans Python and Node/JS agent code, lets you pick decisions for risky tools, and writes wrappers plus .tide/ledger/policy.json in one terminal flow. Omit --pack for baseline; add --pack support-refunds for refund agents.

Python-only install:

python3 -m venv .venv
source .venv/bin/activate
pip install -U tide-runtime
tide run --pack support-refunds

For a non-interactive install, use tide run --pack support-refunds --yes.

tide init --dry-run --pack <name> prints the policy before writing it. Built-in packs: baseline, support-refunds, sql-analyst, mcp-baseline.

To have Tide suggest wrappers for an existing Python or Node/JS agent:

tide scan --path . --pack support-refunds
$EDITOR .tide/tide-plan.json
tide apply --interactive --write-policy

--interactive shows each proposed tool wrapper with its file, function, inferred tool name, risk class, captured fields, and runtime decision. Press enter to accept the suggested decision, or type allow, log, escalate, or deny (block) to write an explicit per-tool policy override. Use tide apply --yes --write-policy to accept the reviewed plan non-interactively.

from tide import configure, controlled_tool, record_outcome

configure(agent_id="crm-agent", workflow_id="lead-sync", environment="prod")

@controlled_tool(name="crm.update_lead", risk_class="internal_write", capture=["lead_id"])
def update_lead(lead_id, status):
    return {"lead_id": lead_id, "status": status}

try:
    update_lead("L-1", "qualified")
    record_outcome("success")
except Exception:
    record_outcome("failure")
    raise

Default files live in .tide/ledger/. Override with TIDE_LEDGER_DIR or configure(project="...").

tide approvals
tide approve <id>
tide deny <id> --reason "not in scope"
tide report