IDEN
Book a call

About the studio / 01

Agentive Labs

AI and product engineering studio in Jakarta

Agentive Labs is an AI and product engineering studio based in Jakarta Metropolitan Area. We build systems that run in production: AI agents wired into internal systems, data platforms, and digital products people use every day.

The studio was founded in 2023. We work with companies in Indonesia and with clients abroad, contracting in dollars for international engagements and in rupiah for local ones.

Three stone planes standing above water at twilight
Founded
2023
Based in
Jakarta Metropolitan Area
Contract currency
USD / IDR

02 / Engineering stack

The technology we assemble

We choose components for the workflow, data, and risk at hand, rather than putting every framework in every project.

Business sources

01Documents
02Internal systems
03Human decisions
Agent layer

Work outputs

01Approval
02Digital product
03Traces & insight
One architecture pattern: business sources → permissioned agents → reviewable actions.
01

Agents & orchestration

Agents use tools inside permission boundaries with typed outputs.

  • Pydantic AI
  • Mastra
  • Google ADK
  • MCP
  • LangChain
  • LlamaIndex
02

Data & retrieval

Knowledge and transactions retain traceable sources.

  • PostgreSQL
  • Pinecone
  • Redis
  • BigQuery
  • Kafka
03

Product & APIs

Interfaces and services are built around daily work.

  • TypeScript
  • Python
  • Go
  • Next.js
  • FastAPI
04

Quality & operations

Evaluation, execution traces, and costs remain visible after release.

  • Langfuse
  • OpenTelemetry
  • Sentry
05

Cloud & deployment

Infrastructure and release workflows are chosen to fit the product.

  • Google Cloud
  • AWS
  • Vercel
  • Cloudflare
  • GitHub Actions

03 / Applied intelligence

From bottleneck to working system

We connect data, permissions, and evaluation into workflows people can use. We measure the effect together after release rather than promise it upfront.

  1. Scattered knowledge becomes working context

    Documents and operational data are indexed with clear permissions and sources. Hybrid search and RAG give an agent relevant context before it drafts an answer or next step.

    Hybrid search · pgvector · sourced RAG

  2. Repeated tasks become controlled actions

    Agents use tool calling and bounded APIs to read, prepare, or perform tasks. Outputs are validated; actions affecting money or customers wait for human approval.

    MCP · typed outputs · human-in-the-loop

  3. System behavior is visible, not guessed

    Real examples become an evaluation dataset. Tool calls, failures, latency, and model cost are tracked so each change can be compared with the previous version.

    Evals · Langfuse · OpenTelemetry · cost per task

04 / From Jakarta

Where we work

  • Jakarta
  • Indonesia
  • Southeast Asia
  • Global (remote)

Jakarta time (WIB, UTC+7). For clients in other time zones we schedule calls in the overlapping hours.

05 / About the studio

Have something to build?

Tell us the problem. If it is not work for us, we will say so.

To get started

A rough outline is enough.

  1. The problem to solve
  2. Systems it touches
  3. A useful first result
Discuss a projectSee our work