Skip to main content
1

Write a research prompt with a consistent template

The key to useful parallel research is giving every session the same checklist. Each session researches one library independently, so the template ensures results are directly comparable when merged.Open a new Devin session from the Devin home page, or use the Explore Advanced Capabilities page on the Devin home page for a parallel research prompt template.
2

Review and approve the proposed sessions

After submitting, Devin parses your list and proposes one session per library. You’ll see a preview like:
Proposed sessions (30):
  1. Research Datadog Logs — pricing, SDKs, retention, alerting...
  2. Research Grafana Loki — pricing, SDKs, retention, alerting...
  3. Research AWS CloudWatch Logs — pricing, SDKs, retention, alerting...
  ...
Review the list and click Approve to launch all sessions simultaneously. Each session runs independently — browsing the library’s website, reading documentation, checking developer forums, and filling in the template.If you want to skip or add libraries, edit the list before approving. You can also attach a playbook to ensure every session follows the same research methodology.
3

Collect and compare results

Once all sessions complete, Devin automatically merges the individual reports into a single comparison. The output follows whatever format you requested — here’s what the compiled spreadsheet-style comparison looks like:
## Logging Library Comparison (Node.js + Python, 2 TB/day)

| Library           | Type       | $/mo (2 TB/day) | Retention       | Node SDK | Python SDK | Query Lang   | Alerting     |
|-------------------|------------|-----------------|-----------------|----------|------------|--------------|--------------|
| Datadog Logs      | SaaS       | ~$5,400         | 15d hot, archive| 5/5      | 5/5        | Custom DSL   | Yes + anomaly|
| Grafana Loki      | Self-host  | Infra only      | Configurable    | 4/5      | 4/5        | LogQL        | Via Grafana  |
| Axiom              | SaaS       | ~$1,200         | 30d hot, 1yr    | 4/5      | 4/5        | APL          | Yes          |
| Better Stack      | SaaS       | ~$890           | 30d default     | 5/5      | 4/5        | SQL-like     | Yes          |
| Elastic Cloud     | SaaS/self  | ~$3,600         | ILM policies    | 5/5      | 5/5        | KQL / Lucene | Yes + ML     |
| Signoz            | Self-host  | Infra only      | Configurable    | 4/5      | 4/5        | ClickHouse SQL| Yes         |
| Coralogix         | SaaS       | ~$2,100         | Hot/warm/cold   | 4/5      | 3/5        | Lucene / SQL | Yes + anomaly|
| ...               |            |                 |                 |          |            |              |              |

### Top 3 for a 50-service Node.js + Python stack:
1. Axiom — lowest cost at scale, fast APL queries, solid SDKs
2. Grafana Loki — zero license cost, pairs with existing Grafana dashboards
3. Datadog Logs — best SDK auto-instrumentation, but expensive at 2 TB/day
You can ask follow-up questions in the same session — it has context from all the child sessions.Once you’ve picked a winner, you can launch a Devin session directly from the same session to set up the library in your repo:
4

Go deeper on the shortlist

Once you have a shortlist, start targeted follow-up sessions for deeper evaluation.
5

Tips

This pattern works for any technical evaluation

Parallel research isn’t limited to logging tools. Use it for any evaluation where you need the same data points about many options — CI/CD platforms, feature flag services, ORMs, cloud providers, or compliance frameworks. Example: “Research these 20 CI/CD platforms and compare build speed, pricing, self-hosted options, and GitHub integration quality.”

Keep each session scoped to 15-30 minutes

If a single library needs hours of deep investigation, that’s a sign it should be its own focused session rather than part of a parallel run. Parallel sessions work best when each item takes roughly the same amount of effort.