Custom AI agent development

Turn repetitive documents and workflows
into a custom AI agent

Data retrieval, document Q&A, spreadsheet processing, file organization, process automation — we turn repetitive, document- and process-heavy tasks into runnable, testable, deliverable agents. We assess whether it can be done first, and build only what's doable.

Tell me your scenario and I'll help frame the requirement.Xiaoju, the requirements assistant
What we can build

Custom agents, built around your workflow

These are common directions we can build to order. Whether and how each can be done, we assess feasibility first — we build what's doable and tell you honestly what isn't.

Data retrieval agentUnderstands a data need, runs a real search on authoritative platforms, returns source links and an availability note.
Document & knowledge-base Q&A (RAG)Turn internal materials, documents and policies into a knowledge assistant whose answers cite their sources.
Spreadsheet processing agentBatch-clean Excel / CSV, unify field definitions, aggregate and compute by rule.
File organization agentClassify, rename, archive and preview local files; actions are previewable and reversible.
Data collection & extractionCollect, clean and structure public data by rule — within compliance limits, page by page.
Process automation agentHand a repetitive, tedious data workflow over to an agent that runs it step by step.
Research analysis & organizationOrganize literature and panel data into standardized tables with notes on field definitions.
Compliance & industry knowledgeTurn public regulations, cases and domain materials into searchable, structured knowledge.
What you get

More than a chat box

A custom build delivers the items below as needed (exact scope confirmed with you).

Requirement specGoals, scope, inputs/outputs and boundaries, written down.
Prompt & workflow designHow the agent reasons, calls tools and handles failure.
Knowledge-base preparationYour materials organized so they can be retrieved and cited.
Tool & permission configWhich tools it may call, and where the boundaries are.
A usable page or chat entryA page or chat entry you can use directly.
API or automation scriptsWhen system integration is needed, an interface or scripts.
Test question setA set of real questions and expected results for acceptance.
Docs + acceptance reportHow to use it, and the acceptance results, written up.
Maintenance planHow it's maintained and iterated after launch.
How we sign off

What counts as done

Acceptance criteria are agreed before we start, then checked point by point.

Completes the agreed tasksGives usable results for the agreed tasks.
Cites the specified sourcesAnswers can cite the materials you specify.
Handles the agreed formatsProcesses the agreed file / data formats.
Results are exportableOutput can be exported in the form you need.
Fails loudlyWhen it can't, it says so — no pretending.
Has permission limitsClear boundaries on what it can and can't do.
TraceableKey actions are logged and reviewable.
How we work

Five steps, from need to launch

Assess first, build second; whether it can be done is answered honestly at step 2.

1 · Scope the needClarify the business goal, materials and data situation.
2 · Feasibility checkJudge whether and how far it can be done — said honestly.
3 · Plan & quoteA delivery plan, deliverables and a quote.
4 · Deliver & acceptDeliver to scope, then check point by point.
5 · Maintain & iterateOngoing maintenance and improvement after launch.
Proof of capability

See the agents we run ourselves

These are our own agents and tools, running today — the same capability goes into your custom build.

Live

Web Xiaoju · requirements & retrieval assistant

Describe the data you're looking for; it organizes the search criteria, runs a real search across authoritative data platforms, and produces a data-availability assessment report. No install or sign-up — just open and try.

In beta

Browser extension · web data assistant

Collect tables, lists and product information from a web page with one click and export to Excel or CSV; data is processed locally throughout, never routed through a server.

In closed beta, opened in batches after you register; we'll tell you how to get the install package when your batch opens.

Coming soon

Desktop · Xiaoju Disk Cleanup

Quickly locate what's using your local disk space, support safe cleanup, and let AI help tell apart files safe to clear from those worth keeping.

Coming soon; registration is open and we'll notify you at launch.

Tell us your scenario — we assess first

Share your goal and your materials; we run a feasibility check, then propose a plan and a quote. Priced per project complexity and scope; the initial consultation is free.

Custom enquiry