DalangRuns on SemarPrivate preview

One model to command them all.

A multi-agent AI system delivered as one model. One endpoint, one answer. Many models, one mastermind.

(01)What is Dalang

One endpoint. Two protocols. No migration.

Dalang coordinates a pool of models behind a single model string, and speaks the API protocols you already use: OpenAI and Anthropic.

Point your existing OpenAI or Anthropic client at Dalang, change base_url and your key, and keep your SDK. The coordination happens behind one answer.

base_url = "https://dalang.thelabs.id/v1"
api_key  = "sk-thelabs-…"

# POST /v1/chat/completions
model    = "semar"
(02)The model

Meet Semar.

One published model, named for the wayang's wise guardian — the humble figure who quietly carries the whole performance.

Semar is a single model string backed by a coordinated pool. Easy requests stay fast and cheap; the hard, multi-step ones get more effort automatically. You never pick a tier.

Read the model docs
semarOne published model

Balanced quality and latency for everyday work, with more effort spent automatically on the hard, multi-step problems.

Good at

Coding & reviewResearchAnalysisAgentic workflows
(03)How it works
01

One API, a coordinated pool

Reach a coordinated pool of models behind a single endpoint and one model string.

02

Superior on complex tasks

Coordination lifts quality where a single model stalls on multi-step work.

03

Flexible, observable control

Tune for balanced speed or maximum quality; see what you need, not the plumbing.

04

Opaque routing by design

Which models run, and how they coordinate, stays internal — the moat, not a setting.

The moat

Coordination, proven by measurement.

The differentiator is not the plumbing. It is the coordination policy: built eval-first and held to evidence.

An eval harness comes before the product. Which models run, and how they are coordinated, stays internal by design. We publish results, not routing tables.

Discipline

Eval harness built first
Claims trace to artifacts
Routing opaque to the caller
Benchmarks: to be published
(04)Use cases

Built for work that takes more than one step.

Coding & review

Plan, implement, and review changes across a codebase — a coordinated pool catching what one pass misses.

prompt

Refactor our auth flow to handle token refresh, and add tests.

semarPlanned the change, implemented it across three files, and added a suite of tests, all passing. The summary and diff are ready for your review.
(05)Pricing

Pay for what you use.

Preview pricing — indicative and subject to change before general availability.

Free

$0/mo

Get started at no cost.

+Semar model
+50 messages / day
+Community support
Start free

Plus

$20/mo

For regular use.

+Everything in Free
+1,000 messages / day
+Higher rate limits
+Email support
Start

Pro

$200/mo

For heavy, daily use.

+Everything in Plus
+5,000 messages / day
+Priority capacity
+Priority support
Start

Need a larger quota, an SLA, or on-prem? Talk to us about Enterprise.

(06)FAQ
Point your existing OpenAI or Anthropic client at the Dalang base URL, swap in your key, and set the model to semar. No SDK migration.
There is one published model: semar. It is backed by a coordinated pool behind the scenes, so you get one model string and one answer, with the routing handled internally.
Both. Dalang speaks the OpenAI (/v1/chat/completions) and Anthropic (/v1/messages) wire protocols. Change base_url and key — keep your SDK.
No — and that is the point. Which models run and how they are coordinated stays internal. The differentiator is the coordination policy, proven by measurement.
See the privacy policy for collection and retention. Enterprise plans add data controls, a DPA, and audit logging.
Two ways. Take a monthly subscription with an included message quota — Free, Plus, or Pro — or pay as you go, metered by token at $3 per million in and $15 per million out. Enterprise is custom. You can switch any time. Preview pricing is indicative and may change before general availability.

Get early access.

Dalang is in private preview. Leave your email and we'll send an invite when a spot opens.

We'll only use your email to send an invite. See our privacy policy.