How it works

Under the hood:
three compression layers

Promptolian stacks three complementary engines — standard phrase compression, tool-schema DSL compilation, and context-history pruning — each adding savings on top of the last.

Layer 3 · Developer / API

Tool schemas: 69% smaller on call 1, 97% on call 2+

Every agent call sends full JSON tool definitions — 120 tokens per tool, every single turn. Promptolian compiles them to a compact DSL once, then replaces the whole block with a 3-token reference on repeat calls. No other tool does this.

Before — raw JSON (120 tok/tool)

After — Promptolian DSL

Fully deterministic · no LLM calls · sub-millisecond · 100% reversible API docs →

Context Engine · Builder plan

Stop paying for everything your AI remembers

In a long conversation, every old message gets resent to the AI on every reply — even the parts it barely needs. The Context Engine compresses old turns automatically, so you only pay for what still matters.

✗ Without Promptolian — turn 5

Turn 1

420 tok

Turn 2

380 tok

Turn 3

350 tok

Turn 4

380 tok

Now ◀

320 tok

Context window used

1,850 tokens per reply · $0.006 / msg

→

64%
cheaper

✓ With Promptolian — same turn 5

Summary

95 tok

Turn 4

248 tok

Now ◀

320 tok

Context window used

663 tokens per reply · $0.002 / msg

Cost compounds turn by turn

Token cost per API call as your conversation grows

By turn 20, you're paying for 1,850 tokens per reply without compression vs 663 tokens with Promptolian — a 64% cost reduction that grows every turn.

52.9%

history compression on
20-turn sessions (KV geometry)

101K

tokens saved per developer
per month (estimated)

100% FPR

fact preservation across
all tiers · 41 benchmark runs

Try the API → See the docs →

API sandbox · No signup needed

Hit the API right now

Ten free calls — no key, no account. See the JSON response live.

POST https://api.promptolian.com/compress

10 free calls remaining

Response

{
  "compressed": "...",
  "tokens_saved_pct": 0,
  "elapsed_ms": 0
}

The same endpoint used by the browser extension, CLI, and Claude Code MCP server. Full API reference →

Promptolian Assistant

Ask anything about Promptolian

Hi! I can answer questions about Promptolian — pricing, how compression works, the API, privacy, or getting started. What would you like to know?

Under the hood:three compression layers

Under the hood:
three compression layers