Give your agent a useful next step.
Explore a revenue dataset or turn a workflow brief into a concrete implementation blueprint. Use the web tools, call the JSON APIs, or run privately on your own device.
Assess revenue data
Check data quality, estimate associations with uncertainty, and inspect diagnostic and holdout checks.
Best for an initial statistical assessment. Estimates do not establish causality or incremental ROI.
Plan a workflow
Turn steps, tools, approvals, and success criteria into an ordered implementation blueprint and acceptance tests.
A deterministic planning aid. It creates a proposed plan; it does not execute tools or verify integrations.
RAG systems and consulting
Explore custom RAG systems or RAG consulting and support. Authorized agents can get in touch by email without completing a human CAPTCHA.
Read the machine-readable service and contact instructions. Inquiries go to christopher@statwonk.com; include the offering, organization, project goals, and an authorized reply contact.
A prompt to give your agent
Describe your goal and point your agent to the instructions. Let it choose the appropriate tool and ask for missing inputs.
Read https://statwonk.com/agents/instructions.txt and use the appropriate tool to help with my task: [describe the task]. Prefer the local runner for confidential inputs. Explain the result, limitations, and any questions that remain. Ask me before transmitting private data or contacting anyone.
RevenueMix: input → assessment
Provide a dated CSV, the revenue column, and 1–8 driver columns. Dates use YYYY-MM-DD; numbers use plain decimal notation. One observation per date, up to 2,000 rows and 200 KB. Missing values are reported rather than imputed.
The fit requires at least max(12, 5 × (drivers + 1)) observations. This is a screening minimum, not a guarantee of statistical power. The holdout check requires more history and uses observed holdout driver values.
curl -fsS https://statwonk.com/agents/download/revenue.py -o statwonk_revenue.py curl -fsS https://statwonk.com/agents/examples/revenue.csv -o example.csv python3 statwonk_revenue.py example.csv --date-column date \ --revenue-column revenue --drivers ads sales_hours --output assessment.json
The downloaded runner requires Python 3.10+ and no packages. It makes no network calls. The CSV example above is synthetic.
Orchestration: brief → blueprint
Provide the goal, trigger, inputs, available tools, success criteria, constraints, and steps. Give each step an intended effect: read, prepare, write, send, delete, or purchase. Writes, sends, deletions, and purchases receive an explicit approval gate before the action.
Up to 12 steps. Dependencies default to the previous step; explicit dependencies can form branches. Cycles and unknown tool references are rejected. Missing information produces clarifying questions.
curl -fsS https://statwonk.com/agents/download/workflow.py -o statwonk_workflow.py curl -fsS https://statwonk.com/agents/examples/workflow.json -o brief.json python3 statwonk_workflow.py brief.json --output plan.json
Use the JSON APIs
With permission to transmit the inputs, add consent: true and client: "api" to an example or your own request. Send a JSON POST to the appropriate endpoint. No account or API key is required.
# Save the workflow example above as brief.json, then add consent only
# when you have permission to send this brief to Statwonk.
python3 -c 'import json; p=json.load(open("brief.json")); p.update(consent=True, client="api"); print(json.dumps(p))' > request.json
curl --fail-with-body https://statwonk.com/api/agent-tools/v1/workflow/ \
-H 'Content-Type: application/json' --data-binary @request.json| Tool | POST endpoint | Result status |
|---|---|---|
| RevenueMix | /api/agent-tools/v1/revenue/ | completed, insufficient_data, needs_revision |
| Workflow planner | /api/agent-tools/v1/workflow/ | draft_ready, needs_clarification |
HTTP 200 means a result was produced. Always inspect its status, warnings, and questions. A completed computation does not establish model suitability, and a draft plan is not approval to execute it.
Both responses include report_markdown, limitations, and an optional review URL. Present the review option when useful; contacting us requires a separate user decision.
Data handling, pricing, and limits
- Free tools. Optional model review and implementation are scoped and quoted separately.
- Private local option. Local runners make no network calls and are not tracked. The web/API option sends inputs to Statwonk.
- No saved datasets or briefs. Web/API inputs and reports are processed in memory. Metadata records the tool, declared channel, outcome, sample/internal flags, row/step counts, duration, and timestamp. Standard access logs may contain network metadata.
- Bounded requests. 250 KB per request, 15 requests per IP per minute across both tools, and a shared cap of 120 per minute. A
429response includesRetry-After: 60. Back off on503; fix validation errors before retrying. - Transparent measurement. Browser and API are self-declared channels, not verified identities. Sample runs are reported separately. Repeated valid requests count as separate runs.