README
¶
squoze
Your context, squoze.
Universal, deterministic LLM context optimizer. One static binary that sits between your AI agent and any OpenAI/Anthropic-compatible provider and shrinks request bodies — without losing information the model needs, and without breaking provider prompt caches.
squoze detects the wire format, classifies content blocks and squeezes verbose machine output (test runs, logs) with head/tail elision that never drops error lines. Compression presets adapt per model family, and every elision is locally reversible by ref.
Status: stable, v0.2.0 (Squoze v2). Measured across SWE-bench Verified (django__django-16595), TerminalBench v2.1, Aider Polyglot, and Live GitHub Pull Requests (go-chi/chi#1171): 100% Pass@1, <0.6 ms streaming latency, and up to 98.9% byte savings across test logs, dependency lockfiles (pnpm-lock.yaml, go.sum), and repetitive tool reads. Full protocol and gates: docs/eval-protocol.md.
Why
Agents waste most of their token budget on noise: repeated file reads, verbose CLI output, stale tool results. squoze removes the noise and keeps the signal — deterministically, with quality contracts:
- Fail-open — any error means your request goes through untouched.
- Never-elide — errors, failures and stack traces are never compressed away.
- Cache-safe — decisions are stable across turns so provider prompt caches keep hitting (a 90% discount is worth more than a 20% squeeze).
- Reversible — originals stay local; the model can pull them back on demand.
Install
go install github.com/Rethinger/squoze/cmd/squoze@latest
No prebuilt binaries yet — go install or go build from source. On Windows a
freshly built binary may trip a Defender false positive (ThreatID 251873); add
an exclusion or sign it locally.
Usage
# Drop-in proxy for any OpenAI/Anthropic-compatible client
squoze proxy --port 8787 --upstream https://api.anthropic.com
ANTHROPIC_BASE_URL=http://localhost:8787 claude
# Or zero-config: wrap the agent, env injection does the rest
squoze wrap --upstream https://api.anthropic.com claude
# Resolve an elision marker ref back to the full original text
squoze retrieve a3f9c2e1b4d0
# Check savings against a live provider, or patch opencode's provider list
squoze livecheck
squoze oc
Every response carries X-Squoze-Original-Bytes, X-Squoze-Sent-Bytes and
X-Squoze-Format.
Use as a library
The root package is a thin facade over the engine — one instance per process keeps the decision memo warm, which is what makes repeat turns byte-identical:
import "github.com/Rethinger/squoze"
eng := squoze.NewEngine(squoze.DefaultMemoCapacity)
out, res := eng.Apply(requestBody)
This is how 2papi embeds it as an optimization mode.
Build
go build ./...
go test ./... # unit + quality contracts
go test ./internal/eval -v # was→is savings table
Apache-2.0. Pending work is tracked honestly in docs/UNFINISHED.md.
Documentation
¶
Overview ¶
Package squoze is the library facade for the universal, deterministic LLM context optimizer ("your context, squoze.").
Typical gateway embedding:
eng := squoze.NewEngine(squoze.DefaultMemoCapacity) out, res := eng.Apply(requestBody)
One Engine instance per process keeps the decision memo warm: identical original bytes always produce byte-identical output, which keeps provider prompt caches stable across turns. See README.md for the quality contracts (fail-open, never-elide, cache-safe, reversible).
Index ¶
Constants ¶
const DefaultMemoCapacity = engine.DefaultMemoCapacity
DefaultMemoCapacity bounds the per-process decision memo (~4k blobs).
const Version = engine.Version
Version mirrors the CLI version.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
Source Files
Directories
¶
| Path | Synopsis |
|---|---|
|
cmd
|
|
|
squoze
command
Command squoze is the universal, deterministic LLM context optimizer.
|
Command squoze is the universal, deterministic LLM context optimizer. |
|
internal
|
|
|
compress
Package compress implements the first real transform: head/tail elision of verbose machine output, under the quality contracts.
|
Package compress implements the first real transform: head/tail elision of verbose machine output, under the quality contracts. |
|
engine
Package engine hosts the squoze optimization pipeline.
|
Package engine hosts the squoze optimization pipeline. |
|
eval
Package eval hosts the deterministic quality-fixture harness: realistic tool-output classes with must-survive facts and minimum savings floors.
|
Package eval hosts the deterministic quality-fixture harness: realistic tool-output classes with must-survive facts and minimum savings floors. |
|
harness
Agent-level presets: the coding agents themselves (as opposed to raw providers).
|
Agent-level presets: the coding agents themselves (as opposed to raw providers). |
|
profile
Package profile adapts compression aggressiveness to the target model family.
|
Package profile adapts compression aggressiveness to the target model family. |
|
proxy
Package proxy implements the drop-in HTTP proxy mode:
|
Package proxy implements the drop-in HTTP proxy mode: |
|
router
Package router classifies text blobs so the engine can decide WHO gets squeezed and who must pass through untouched.
|
Package router classifies text blobs so the engine can decide WHO gets squeezed and who must pass through untouched. |
|
store
Package store provides the content-addressed decision memo behind the cache-guard contract.
|
Package store provides the content-addressed decision memo behind the cache-guard contract. |
|
wire
Package wire detects the wire format of an LLM request body.
|
Package wire detects the wire format of an LLM request body. |
|
wrap
Package wrap implements `squoze wrap CMD`: run an agent command with its provider base URLs pointed at an in-process squoze proxy.
|
Package wrap implements `squoze wrap CMD`: run an agent command with its provider base URLs pointed at an in-process squoze proxy. |