Public Tools / Try Now

Free tools.
Public code.

Two open-source tools you can run today. Both ship the public-tier reduction — honest numbers, honest mechanism, no gate. Paid tiers deliver what's sealed.

Conduit Reduction Desktop widget

Paste text, get reduced text back. WPF desktop app, runs offline. MIT.

~40–50% reduction GitHub →
Conduit Open Python middleware

Drop-in middleware for any OpenAI-compatible endpoint. FastAPI + CLI. MIT.

Same algorithm, server-side GitHub →
Request AccessRead the docs →

Context Intelligence / V1

AI that only carries
what it needs.

Conduit is a trust-mediated context broker that eliminates the carrying cost of AI memory — delivering verified context, at the right moment, with cryptographic proof of every source.

Up to 50%
Context reduction
25M tok
Corpus size
<1s
Recall latency
Reduction
Corpus reduction Sealed
Token reduction Up to 50%
Availability
Public Open
Enterprise Sealed
Gov Sealed
Partner Sealed
Request AccessView proof docs →

App-to-App Context / V1

Any app.
One context.

Your IDE, browser, terminal, and AI share a single live context layer without any app knowing the others exist.

Unlimited
Live connections
Real-time
Cross-process sync
Zero
App modifications
Request AccessLearn more →

Reduction Pipeline / V1

Every stage.
One receipt.

Multi-stage reduction pipeline compounds context savings at every layer. Every step is auditable. Enterprise tiers sealed.

Multi-stage
Pipeline
Up to 50%
Public reduction
Near-zero
Cost per query
Reduction
Corpus reduction Sealed
Token reduction Up to 50%
Availability
Public Open
Enterprise Sealed
Gov Sealed
Partner Sealed
Request AccessSee proof →

Cryptographic Proof / V1

Every answer
has a receipt.

Every answer is labeled and sealed at the source. FROM CONTEXT, FROM TRAINING, FABRICATED — independently verifiable. The model cannot hide where its answer came from.

Sealed
Every emission
3
Source labels
Verified
Audit coverage
Request AccessVerify a proof →
01 / 05
public release/ open source
MIT
conduit-reduction Desktop
WPF widget · paste / reduce / copy · offline
~40–50% reduction github →
conduit-open Middleware
Python · FastAPI + CLI · OpenAI-compat
Same algorithm, server-side github →
Public tools ship the public tier. Enterprise, Gov, Partner tiers sealed.
conduit/ claude-sonnet-4-5
Local mode
○ Context filter — 2 sources matched · 9,100 → 5,000 tok−45%
● From Context
Every context emission is sealed and independently verifiable. Each answer carries a proof of its source.
What is the capital of France?
○ No match — routing to model knowledgeTraining
● From Training
Paris.
Ask anything…
↑
strands/ live cord system
Live
Connected applications
📄
Visual Studio Code
conduit-pipeline.cs · 2,341 chars
live
🌐
Chrome — docs.anthropic.com
Page context · 1,180 chars
live
⌨
Terminal — zsh
Last 40 lines · 890 chars
live
Combined context
Total available4,411 chars
Sent this turnActive
Ask about any connected app…
↑
reduction pipeline/ multi-stage
Cloud mode
○Structural pass — payload compressionReduced
○Entity pass — memory graph compressionReduced
○Reference pass — conversation history compressionReduced
○Residual pass — final byte compressionReduced
25M tok
Corpus size
Verified
Output
Up to 50%
Reduction
proof system/ v1
Sealed
corpusllama.cpp · 25M tok
sentVerified
reductionVerified
sourceFROM CONTEXT
proofsealed · verifiable
● From Contextanswer derived from corpus entries this turn
● From Traininganswer from model knowledge, no corpus match
● Fabricatedmodel is uncertain — treat as unverified
Request early access

We'll send an access link as soon as a slot opens.

No spam. Just an access link when you're up.

You're on the list.

We'll be in touch.