CUSTOM AI AGENTS · Research & data
Knowledge researcheras part of your team.
Search documents. Compare versions. Answer with source references.
Explore this agent for my team ↗01 · THE WORK, STEP BY STEP
How the task moves forward.
Information helps when it can be checked. We define questions, periods and sources for analysis that explains both findings and limitations.
- 01
Search documents
Searches only spaces the agent is authorized to access.
- 02
Compare versions
Flags conflicting versions and separates current documents from historical material.
- 03
Answer with source references
Answers with locatable references and states when evidence is insufficient.
02 · ILLUSTRATIVE EXAMPLE
A situation that may feel familiar.
An engineering firm finds approved specifications from an earlier project.
Illustrative scenario, not a customer case study or measured results.
SlackExampleRead about the scope of the assignment at the team's channel.
GitHubExampleUse one route consistent with the trace and example implementation.
Export contract
| Element | Documented behavior |
|---|---|
| Route | GET /api/orders/export |
| Permission | orders:export |
| Account | Retrieved from the session |
| Filter | only orders paid |
| Response | CSV: date, client and total |
| Access denied | HTTP 403 |
NotionExampleDocument the request and output with fictional data and an explanation of the error.
Export paid orders · Draft
Request
GET /api/orders/export Requires a session and orders:export permission. The account comes from the session; the request cannot select a different account.
Examples CSV · Fictitious data
fecha,cliente,total 2026-10-01,Cliente de ejemplo,29.00
If you get a 403
Ask the owner to check the export permission and account access. Do not change identifiers to try to access another account.
NotionExampleKeep the document in its workspace rather than creating an unnecessary PDF.
- Document
- Export guide
- Status
- Draft · Awaiting technical review
- Examples
- Request, CSV and HTTP 403
SlackExampleShare the result and distinguish prepared work from executed work.
03 · YOUR WAY OF WORKING
Fits the business you already have.
Context and tools
We start with your current documents, systems and channels. We assess feasible connections and define the permissions required for this specific task.
Improves with your feedback
Approved corrections are incorporated into instructions, examples or knowledge sources. We review changes before expanding the agent’s scope.
A clear way to evaluate
We agree on a representative sample and measure review time, errors, exceptions and output quality against the current workflow. Feasibility is decided using that evidence.
04 · COMMON QUESTIONS
Before getting started.
What information does this agent need?
Internal question and permissioned repositories. We review quality, currency and access before connecting the workflow. You do not need to replace all your tools to begin.
What will my team receive?
Traceable answer and reference documents. We agree on the format and delivery channel during setup, together with cases that need review.
Which decisions remain with people?
Conclusions are reviewed in light of data quality, coverage and uncertainty.
Can it coordinate with other agents?
Yes. We define what this agent hands to the next one and who handles exceptions. It can interact through Slack, WhatsApp, email or Microsoft Teams, depending on the integration and permissions we configure.
INDUSTRY APPLICATIONS
Where it could fit.
RELATED AGENTS
A team that works together.
Market researcher
Market report with references and limitations.
See tasks and example ↗Research & dataData cleanup assistant
Reviewable changes and an issue log.
See tasks and example ↗Research & dataReport analyst
Analytical summary with sources and open questions.
See tasks and example ↗START WITH ONE TASK
How would this fit your business?
Tell us how you do it today. We will define the scope, tools and controls with you.