Long sessions compound
As a task continues, every new step can carry a larger and more expensive working history.
Built for expensive, long-running AI work
Powerful AI is expensive. Long tasks multiply the cost.
Longweis Relay helps you keep the premium model and workflow you trust while reducing the model compute required to carry long, continuous and history-rich work to completion.
macOS 13 or later · Apple silicon · Open registration · $10 Relay credit for a new account · optional application adds $50 after matching
Open testing is available now
Best fit: a Mac developer with one repeatable, non-sensitive OpenAI-compatible or agent-runtime workflow to compare before and after Relay.
Optional $50 application bonusTHE RELAY OPPORTUNITY measured usage
50%
less premium-model computeup to, on suitable everyday work
90%
less total model computeup to, on suitable long-memory work
Upper ranges, not guarantees. Actual results depend on the task, model and length of the working history.
The problem Relay is built for
Premium-model pricing hurts most when the work refuses to stay short.
A quick prompt is easy to budget. A real project keeps growing: more turns, more history and more expensive model compute just to stay oriented. Relay is designed for that compounding cost.
Context gets paid for again
History-rich work can repeatedly consume premium capacity even when only a small part matters now.
Replacing the model is not the answer
A cheaper model may lower the bill, but it can also give up the capability you selected the premium model for.
RELAY'S ANSWER
Keep your chosen premium connection. Add a measured optimization layer. Turn on prepared memory when the work needs persistent history.
Trust before traffic
Verify the boundary before you route real work.
You should not have to trust a savings claim or an unfamiliar API blindly. Relay makes the local, hosted, and provider boundaries explicit—and recommends proving value on a non-sensitive task first.
Stays in macOS Keychain
Relay does not send your premium-provider API key to Longweis. The Mac app uses it locally when your selected premium model is needed.
Your tools connect to 127.0.0.1
OpenAI-compatible tools talk to a device-only local endpoint and key. Task content still travels over TLS to Relay and the model services required to process it.
No import is required to start
Compute Optimization works without a memory import. Memory Retrieval is optional and sends only the files or folders you explicitly select.
Signed, notarized, and checkable
The Mac app is Developer ID signed and Apple notarized. Every release publishes the exact SHA-256 checksum so you can verify the download yourself.
A safer first test
Prove one workflow before expanding trust.
The useful metric is total cost per successful task—not a token percentage in isolation.
- 01Choose a low-risk task
Start with a repeatable, non-sensitive task. Do not use secrets, regulated records, or irreplaceable production data during open testing.
- 02Run it both ways
Record the direct result first, then run the same task through Relay with the same premium model and acceptance criteria.
- 03Judge the completed task
Compare success, quality, retries, total tokens, your provider charge, and the Relay charge. Stop using Relay if the combined result is not better for that workflow.
Read the full plain-language boundaries in the open-testing policy.
Your existing connection
Use the premium provider and model you already have.
General API keeps your provider key in macOS Keychain and includes Compute Optimization, account usage, balance, and open-testing pricing.
Bring your premium-provider API
Select the provider and exact model you already use. Its key stays in macOS Keychain.
The result you can measure
Spend less premium compute—not less attention on the result.
Choose the level that fits the work. Both modes are included, and both report measured Relay usage. Percentages are workload-dependent upper ranges, not guarantees.
Compute optimization
Up to about 50%less model compute$0.20 / 1M input tokens$0.80 / 1M output tokens
For everyday OpenAI-compatible tasks that do not need a prepared memory collection.
Memory retrieval + compute
Up to about 90%less model compute$0.30 / 1M input tokens$1.20 / 1M output tokens
Prepare selected local material for long, continuous, history-rich work without repeatedly loading the whole archive.
Open-testing pricing. Relay charges only its own measured model usage. Your premium provider bills your premium-model usage separately. Initial memory preparation has no separate setup charge. Each newly registered account receives a one-time $10 Relay service credit. An optional valid application using the registered email adds another one-time $50 after account matching.
Illustrative combined-cost example
What the two charges look like together.
Example only: illustrative premium-model text pricing of $4 per 1M input tokens and $20 per 1M output tokens. The equations combine the remaining premium-model portion with the published Relay rate.
ModeInput / 1M tokensOutput / 1M tokens
Save48%Up toCompute optimizationUp to about 50% less premium-model compute
$4 × 50% + $0.20 × 50% = $2.10$20 × 50% + $0.80 × 50% = $10.40Input: about 47.5% lower · Output: about 48% lowerSave85%Up toMemory retrieval + computeUp to about 90% less total model compute on suitable long-memory work
$4 × 10% + $0.30 × 90% = $0.67$20 × 10% + $1.20 × 90% = $3.08Input: about 83.3% lower · Output: about 84.6% lowerThis is an upper-range illustration, not a promise or quote. Actual combined cost depends on the selected premium model, its current price, the task, the usable history, and measured Relay usage. Media-generation requests do not go through Relay optimization.
From download to first request
Four steps. One local endpoint.
Your premium-provider key stays in macOS Keychain on this Mac.
Install Relay
Download the signed and notarized Mac app, open the DMG, and drag Longweis Relay to Applications.
Register your account
Register with Email or Google—no application, invitation, or activation code required. A new account receives $10 of Relay service credit once.
Choose your connection
Choose the premium provider and model you already use, then add its API key locally.
Use the fused API
Point OpenClaw or any OpenAI-compatible runtime at Relay’s local endpoint and select relay-auto.
The API your tools see
No new workflow to learn.
Start the Local API in the Mac app, copy its local API key, then use these settings in an OpenAI-compatible client. Relay keeps running after the settings window closes; “Start at login” makes it return automatically after you sign in to the Mac.
base_url = "http://127.0.0.1:18777/v1"
model = "relay-auto"
api_key = "copy from the Relay app"Compatibility, verified by real runs
We publish what passed—and label what has not.
Relay exposes a local OpenAI-compatible API, but protocol compatibility alone is not treated as proof. The runtimes below were exercised against Relay with real prompts, tools, files, multiple turns, or concurrent work.
Tested and supportedVerified September 2026
OpenClaw
Short and long tasks, multi-turn work, tool use, concurrency, and recovery paths tested through Relay.
OpenCode 1.18.30
Long prompts, tool calls, four-way concurrency, and cross-turn continuity passed. Parallel workers use isolated state directories.
Aider 0.82.3
Basic calls, real file edits, cross-turn goal retention, and three-way concurrent work passed without rate-limit failures.
AutoGen 0.7.5
Real function-tool execution and three-way concurrent tool turns passed with correct arguments and final results.
Compatibility reflects the tested versions and scenarios above; it is not a guarantee for every release, plugin, provider, model, or workload. New integrations are added only after repeatable tests.
Verified release
Open to Apple-silicon Mac testers.
Longweis Relay P0.69Apple silicon · macOS 13+
Download DMGDownload, register with Email or Google, and start—no application or invitation required.
Verify SHA-256
4c806d0af2cd721293c0d045abf0c955266732c2733839e99126ca612a7ff6a9shasum -a 256 Longweis-Relay-macOS-vP0.69-arm64.dmgOpen testing process
Download, register, and start using Relay.
Registration is open. The application below is optional and only needed if you want the additional $50 service-credit bonus.
- 01
Download Relay
Install the signed and notarized Mac app on an Apple-silicon Mac running macOS 13 or later.
- 02
Register and start
Create an account with Email or Google and receive a one-time $10 Relay service credit. No application or invitation is required.
- 03
Optionally apply for another $50
Submit the optional application with the same registered email. After we match the valid application to your account, we add one additional $50 Relay service credit.
Optional application bonus
Apply with your registered email for another $50.
You do not need this form to download, register, or use Relay. Submit it with the same email as your Relay account. After we match a valid application to the account, we add a one-time $50 Relay service credit.
You bringApple silicon, macOS 13+, and one repeatable real workflow.
You testOne clear, non-sensitive workflow and candid, actionable context.
We provide$10 Relay service credit on new registration, plus $50 after a valid optional application is matched.
Beta credits pay for Relay service, are not cash, and have no cash value and do not cover your own premium-provider charges. The optional application bonus is granted once after email matching.
Before you start
The practical details.
Do I need to apply or wait for an invitation?
No. Download Relay and register with Email or Google to start. There is no invitation or activation-code step.
What does the $10 registration credit pay for?
It pays for Relay’s measured model usage. In General API mode, your premium-model provider continues to bill your own account separately. When the Relay balance runs low, you can add $10, $25, $50, or $100 through Square.
How do I receive the additional $50 credit?
Submit the optional application using the same email as your Relay account. After we match a valid application to the registered account, we add a one-time $50 Relay service credit. It is service credit, not cash, has no cash value, and does not cover your premium-provider charges.
Are the savings guaranteed?
No. The displayed percentages are workload-dependent upper ranges. Short or high-risk tasks may stay mostly on the premium model, while long, continuous, history-rich work has more room to save.
What leaves my Mac?
Your premium-provider API key stays in macOS Keychain and is not sent to Longweis. The task content needed to answer a request is sent over TLS to Relay and the model services needed to process it. If you enable Memory Retrieval, only the material you select is prepared and sent.
Is Relay zero-knowledge or end-to-end encrypted?
No. Relay and the model services needed for a request receive the plaintext required to process it. Use a non-sensitive task first, and do not put secrets or regulated data into open testing.
What is not supported in this release?
Image, video, and audio generation requests do not go through Relay optimization in the current release. They continue through your existing model and runtime.
Which computers are supported?
This release supports Apple-silicon Macs running macOS 13 or later. Windows and Linux desktop clients are still being expanded.