Use cases · Agents

Your agent reads 40 pages to answer one question

Every tool call returns everything it found. The next step needs three lines of it — and you pay for all of it, on every hop.

−97% Of an incident log, root cause still found

13.7k→ 0.4k Tokens reaching the model

What you hand over, what comes back

Your agent sends

Tool results
Logs & traces
API responses
Earlier steps
The current step

As context, or in the messages — both work.

CompresLM

Scores every part against what this step is actually doing, keeps that, forwards one call on your key.

Your agent gets

The answer
Tool calls
Streaming
Token counts

Nothing in your loop changes.

In a chain of agents it compounds

Each hop carries the last one's output forward. By the fourth, most of the prompt is work already done — and billed again.

Researcher

step 1

Analyst

step 2

Reviewer

step 3

Writer

step 4

Prompt as it arrives What the step actually needs

Illustrative shape, not a measurement — how much is removed at each hop depends on your material. The playground gives you the real number.

Connect it

Change the base URL. Keep your SDK, your model, your key.

# Python — OpenAI SDK
client = OpenAI(
    base_url="https://compreslm.com/v1",
    api_key="YOUR_COMPRESLM_TOKEN",
    default_headers={"X-LLM-API-Key": "YOUR_PROVIDER_KEY"},
)

# Tool output goes in `context`, the step stays the message.
resp = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Which service caused the timeouts?"}],
    extra_body={"context": tool_output},
)

Long tool results are the best case there is. See the request shapes for the details.

Run this exact case

A day of incident chatter, one question, both answers side by side.

Try it — no account

Talk to us

Connect it — we'll get it running against your traffic and check the numbers with you.
Fit it to you — closed network, unusual documents, a shape we don't handle yet.
Build it — no product to connect yet? Our partners build it, we make it cheap to run.

Usually a reply the same working day.