This example does a live search. Give it a topic; it finds public sources and writes a Markdown brief with links, supporting quotations, and the limits of what it found. The default topic is battery technology. You do not need an X account or a separate research API key.
Inspect a completed run.
Read the brief from September 10, 2026 and its source dates and check results. These files came from a real local run of the earlier executor. The current template has not had a new live research run. A new run searches a new 72-hour window. The test record includes attempt counts and checks that needed a retry.
Set up your tools.
Follow the SDK quickstart first. You need Node.js 22 or later and the Method CLI, signed in. The Method names its researcher model and its web search, page, and file tools. Read it before you run it. A run stops if search is unavailable.
Download and run the example.
Download the method template and topic input. The input contains only {"topic":"Battery technology"}. There are no supplied findings or fictional source notes.
mkdir battery-research
cd battery-research
curl -fL https://withmethod.ai/examples/research-brief.method -o research-brief.method
curl -fL https://withmethod.ai/examples/research-tools.mjs -o research-tools.mjs
curl -fL https://withmethod.ai/examples/research-inputs.json -o research-inputs.json
method login
method validate research-brief.method
# The Method holds its four tools (research-tools.mjs); agent steps use the
# account's hosted model.
method run research-brief.method --inputs research-inputs.json --workspace . --run-dir .runs/researchThe run performs three tasks and a separate model check for each task. It can take several minutes. The timeout allows up to 15 minutes per operation. It saves files locally; it does not publish or send the brief.
Follow the three steps.
- Search. Fix the window at the run start time and the preceding 72 hours. Search at least three queries covering research, technical announcements, and recent reporting. Save up to eight candidate links and the search record.
- Check the sources. Open the candidate pages and follow links to original papers, lab releases, or company announcements. Check dates and claims. Keep up to five distinct developments, with short quotations and source links. Record excluded or blocked sources.
- Write the brief. Use the accepted findings. State the exact time window, attribute claims to their sources, and explain evidence limits and gaps in the search.
Check what “past 72 hours” means.
The window stays fixed for the whole run. An updated page is not automatically a new publication. A search result’s date is not enough. The method checks the original page or its publisher metadata.
If a source shows only a date, its full possible publication day must fit inside the window. Uncertain dates near the start or end are excluded and explained. This can leave out relevant work whose exact publication time cannot be confirmed.
Read the result and its evidence.
method inspect .runs/research --out inspection.json --include-filesOpen .runs/research/summary.json for the final status and.runs/research/result.json for the brief, source review, and search record. The inspection file includes saved inputs, outputs, and check decisions. Read the source links before relying on a claim. Model checks can miss errors.
If a check stops the run, inspect its reason before trying again. Run method authoring recovery for the recovery steps. Use a new run folder after changing the method or its inputs.
A useful result may contain fewer than five findings. If none qualify, the brief explains what was searched and why sources were excluded. This is a limited search, not proof that no new developments occurred. A company announcement remains a company claim.
Change the topic and run it again.
Change topic in the input file to another technology. Use a new run folder each time; its start time sets the new window. Keep previous results so you can compare sources and check decisions.
method run research-brief.method --inputs research-inputs.json --workspace . --run-dir .runs/research-2This example runs when you start it. It has no built-in schedule.
