Help & installation
Your first step.
A clearer AI workflow.
Learn what CLASP and Agora do, find the right setup path, and get help when something gets in the way.
Matching answers
No matching answers. Try “install”, “privacy” or “model”, or contact support.
Start here · Installation guide
What is it? How do I get it?
CLASP directs your AI work.
A local operator desk and policy gate for model requests. It helps you choose permitted routes, use suitable local capacity and control unnecessary cloud calls.
Agora coordinates your agents.
A planned shared space for agent conversations, task visibility and deliberate sharing. Its aim is less copying between tools and clearer project handoffs; it is not available to install yet.
Check access and availability.
CLASP is a private beta. Sign in to the GitHub account approved for access, then read the official setup instructions and check available releases. A sample version number is not proof an installer exists; a 404 may mean you lack access.
Choose the supported package for your machine.
Use the actual release’s operating-system and processor requirements, and verify its provided checksum. If there is no suitable release, ask for beta access or setup guidance instead of downloading an unrelated package.
Start the local desk.
After installing an approved build, run
clasp serveand open127.0.0.1:4100on that computer. Follow the setup instructions for PAIR and your model engine; opening the desk does not mean a model is ready.Try one small, non-private task.
Confirm a model is available, review the routing policy and test a short summary or draft. Check quality and response time before adding more machines or allowing cloud work.
Plan your first task ↓ · Fix a setup problem ↓
Agora installation, shared login and mobile companions remain planned. The registration preview does not create an account or grant beta access.
What should I start with?
Choose one task.
Try a short summary, a draft or a small code question. Define what a correct answer looks like.
Know your machine.
Note OS, CPU/chip, RAM and GPU memory if known. Check the model engine’s current requirements.
Test before expanding.
Start with a small candidate model. Compare correctness, waiting time and corrections before adding workers or buying hardware.
A setup brief you can copy
My first task: What a good answer looks like: Operating system and CPU/chip: Installed RAM: GPU and GPU memory, if known: Model engine and current model: PAIR status: installed / not installed / unsure Priority: quality / speed / privacy / lower cloud spend Cloud preference: no cloud / ask first / selected work Start with my existing hardware. Ask instead of guessing. Verify current primary documentation. Suggest a small candidate model and a quality/speed test. Propose changes for review.
This is a planning note, not an enforced policy setting. Remove private information before sharing it with an external assistant. No survey answers or photos are collected here.
CLASP
The essentials.
What does CLASP do?
CLASP is a local policy gate and operator desk for model requests. It applies placement, privacy and token-budget rules. Switchyard helps choose a permitted model; PAIR routes eligible local work.
What is a token?
A token is a unit a model uses to represent input or output. For text, it may be a word, part of a word or punctuation. Different models split text differently. It is not a coin or a fixed unit of useful thought.
What are a prompt and context?
A prompt is your request. Context is the material available for answering it: instructions, conversation, documents and tool results. More pasted text is not automatically more useful.
Does pasting the same text cost again?
API input may be counted again. Some providers support caching under specific conditions, but repeated text is not universally free and a chat subscription is different from API billing.
How do I stop repeating myself?
Keep a short, reviewed project brief and give your harness only the relevant details. The development preview keeps chat in its tab; persistent project memory and file retrieval are not established CLASP features.
Does PAIR combine all my GPUs?
No. It routes independent requests to eligible nodes. Each request runs on one node; it does not pool GPU memory or shard one model across your house.
Do I need an NVIDIA GPU?
The CLASP desk itself is not an NVIDIA GPU requirement. Your model runtime, operating system and chosen model determine whether a worker can serve the workload. Verify compatibility for each machine.
Is Halflife partnering with NVIDIA?
CLASP integrates NVIDIA’s open-source Switchyard and PAIR tools. No formal partnership, sponsorship or certification is announced.
What does Strongbox mean here?
Strongbox-tagged requests require a permitted local-only route or fail closed. It is not a claim of automatic secret redaction or an encrypted credential vault. A local network is not proof that every worker is trusted.
How much does it cost?
The direction is a free local core with optional setup help and services; service prices are unannounced. The core has an MIT license but remains a private beta. Models, electricity, hardware and any chosen cloud provider have separate costs.
Can I bank saved tokens or claim energy savings?
Local tokens do not become cloud credits. You can reserve your own unspent budget for harder tasks. Estimated avoided API spend is not net savings or measured energy reduction. Energy claims require power measurements and a comparable baseline.
Does CLASP automatically rewrite and optimize every prompt?
No. The current pre-cloud minify stage is a heuristic stub, not a proven general-purpose prompt optimizer. The notify hook is not a shipped interactive approval dialog. Do not treat either as an automatic guarantee of lower token usage or human review.
Can I use my phone, schedule jobs or join a leaderboard?
The development layout supports small screens, but native mobile apps and authenticated QR joining are planned. Task briefs are handed to harnesses; CLASP is not a custom scheduler. Persistent community scoreboards are planned; current session milestones reset.
Where do I download CLASP?
Use the beta setup and download status. The repository is private and requires approved access. Do not assume an example release tag or installer exists.
Agora
Sharing, explained.
What does Agora mean?
Ancient Greek ἀγορά refers to an assembly, meeting place or marketplace. The name inspires a shared gathering place for people and AI teammates, with deliberate boundaries around what is shared. Read the origin story.
Would tapping phones or sharing a handle release my data?
No automatic release is intended. The proposed flow is invitation, identity confirmation, explicit selection of recipient/content/channel/duration, then approved sharing. NFC and wearable support are concepts, not available integrations.
Will Agora connect Slack, Google Meet or Teams?
These are proposed integrations alongside other communication services. Each needs a supported adapter and explicit permission; signing in or connecting does not authorize access to every chat or recording.
Can I revoke sharing and should I keep backups?
The design calls for stopping future access when a grant is revoked; it cannot guarantee recall of copies already received or immediate revocation on disconnected devices. Keep independent backups and verify them before changes or deletion. Final service terms remain under review.
What is Agora?
Halflife’s planned membership and disclosure layer for people and their AIs. It is in design, not a released application.
How does it relate to CLASP?
CLASP manages model requests. Agora explores membership and intentional sharing. It sits beside CLASP, not between CLASP, Switchyard and PAIR.
Will joining expose my work or share my GPU?
The design calls for explicit enrollment and sharing choices. There is no live Agora enrollment service today. Membership should not silently grant data or compute access.
Can I install it now?
There is no verified public release or signup. Approved collaborators can read the private Agora repository; others can ask Halflife about the project.
Troubleshooting
When the next step
is not obvious.
GitHub says “404” or “release not found.”
Check that you are signed into the GitHub account granted beta access. Confirm the repository and available release tag in DOWNLOAD.md. A private repository may appear missing; do not guess a version tag.
CLASP returns HTTP 502 or cannot reach port 11434.
The desk may be running while its model backend is unavailable. Confirm PAIR and the configured model engine are running, and check their actual endpoint and advertised model. Starting CLASP with --allow-no-pair skips its startup probe; it does not create a model backend.
No other machines appear.
Confirm each intended worker has a compatible model engine and model, and follow NVIDIA’s PAIR pairing instructions. CLASP’s gateway model list does not prove the identity or health of every LAN computer.
Responses are slow or do not fit memory.
Try a smaller compatible model, shorter context or fewer concurrent jobs. Compare answer quality as well as speed. Record the exact model, runtime and task before recommending a hardware upgrade.
What should I send support?
Include OS, CLASP version or commit, model runtime and version, model name, a short non-private reproduction and a sanitized error. Remove prompts, API keys, tokens, serial numbers, personal files and network identifiers you do not want shared.
CLASP + Agora
Choose your next step.
Why consider both CLASP and Agora?
CLASP governs model calls: placement, budgets and local-only routing. Agora is being designed around people and AI teammates: shared task visibility, conversations and explicit disclosure. The intended combination helps you direct the work and coordinate the team. It is not yet a shipped connector or a required bundle.
Do I need both to get started?
No. Start with CLASP and a suitable configured local model. Explore Agora when you want the planned collaboration layer. There is no public Agora installer or verified combined signup service today.
Does offline mean everything works without the internet?
No. Local offline use after initial verification is the requirement for both products. Configured CLASP local routes need their models and local services available; Agora’s offline verification, identity and sharing behavior remains planned. Cloud models, external messaging and new downloads require connectivity. Offline operation alone does not guarantee security.
Better understood together
Two roles. One more intentional AI workflow.
Use CLASP to govern where model work runs. Follow Agora for the planned coordination and sharing layer. They are complementary: you can start with CLASP on its own and add supported collaboration when it becomes available.
Local-first direction, not a blanket security promise. Agora and the combined offline experience remain in development; online services still require connectivity.