Frequently Asked Questions

TernaryPhysics is a local CLI tool that deploys AI agents onto your infrastructure resources (Kubernetes clusters, databases, VMs, cloud accounts, etc.) for investigation and troubleshooting. Think of it as having an experienced SRE who can instantly analyze your systems and help diagnose issues.

Three key differences: (1) It runs 100% locally on your machine—no data leaves your network. (2) Agents are specialized for specific infrastructure types, not general-purpose chatbots. (3) The agents form a mesh—when you deploy multiple agents, they share context and can trace issues across your stack.

TernaryPhysics-7B is our quantized LLM optimized for infrastructure investigation. Each agent also includes the TNN™ (Ternary Neural Network) — our proprietary anomaly detector that learns the specific patterns of YOUR resource. Both models run locally on the resource itself. TNN™ is Patent Pending.

Your infrastructure data stays local. Both AI models (TernaryPhysics-7B and the TNN™ routing/auth weight bank) run on the resource itself. The backend only receives billing metadata: a row per command containing the resource name, agent type, command verb (run/scan/ask), GB processed, duration, and timestamp — never the actual question, command output, logs, or metrics.

We collect billing metadata only — one row per command with the resource name you gave the agent, the agent type, the command verb (run/scan/ask), GB processed, duration, and timestamp. We do NOT receive the question text, command output, logs, metrics, or any data the agent read or produced. Your infrastructure telemetry stays on the host.

You can start using TernaryPhysics immediately with the 1 GB free tier. To continue after that, you'll need to add a payment method. We keep it simple — no complex onboarding, just pay for what you use.

Agents use your existing credentials (environment variables, kubeconfig, .pgpass, etc.). Credentials are read locally and never logged or transmitted. Agents request the minimum permissions needed—read-only by default.

Agents categorize all commands as "safe" (read-only) or "dangerous" (mutations). Dangerous commands require your explicit approval before execution. You always have the final say.

You need Python 3.10+ and about 5GB of disk space for the AI model (4.7 GB model + working space). It runs with 8GB RAM minimum; 16GB lets you run several agents on the same host. Works on macOS, Linux, and Windows (WSL).

Run: pip install tp-ops. On first run, it will download the TernaryPhysics-7B model (~4.7 GB). After that, agents run entirely on your resources.

Yes. After the initial model download, tp-ops works completely offline. The AI runs locally, and agents connect directly to your infrastructure.

TernaryPhysics-7B is optimized specifically for infrastructure investigation and works with the TNN™ anomaly detector. Custom models are not currently supported.

SSH to your resource, install tp-ops, and run "tp-ops drop". The agent auto-detects what type of resource it's on (K8s, Postgres, Redis, VM, etc). For example: ssh prod-db && pip install tp-ops && tp-ops drop

When you deploy multiple agents, they form a mesh network and share context. If your Kubernetes agent finds a pod crashing due to database connection errors, and you have a PostgreSQL agent deployed, they can correlate findings automatically.

Only if you approve them. By default, agents operate in read-only mode. If an agent suggests a remediation (like restarting a pod or killing a query), it will ask for your approval first.

After dropping an agent, use: tp-ops ask <resource-name>. Then just type your question in natural language: "Why is CPU high?" or "Show me the slowest queries".

Agents are priced at $0.50 per GB of data processed. This is a simple, usage-based model—you only pay for what you use. There are no subscription fees, no minimums, and no hidden charges.

Data processed includes logs analyzed, metrics queried, and query results returned by agents. Read-only investigation commands (like checking pod status or listing processes) are included. Large data transfers like full database dumps count toward your usage.

Yes. You get 1 GB of free data processing per month to try out agents on your infrastructure. After that, it's $0.50/GB. Most investigation sessions use a few hundred MB.

You'll be prompted to add a payment method after exceeding the free tier. We accept all major credit cards. Billing is monthly based on actual usage.

The TernaryPhysics-7B model is ~4.7 GB. If the download is slow, check your internet connection. The download only happens once per resource.

TernaryPhysics-7B runs on CPU without requiring a GPU. Response speed depends on your resource's hardware — modern CPUs provide real-time conversational responses. The TNN™ anomaly detector runs in sub-millisecond time.

Make sure your credentials are properly configured (kubectl context, environment variables, etc.). Try running the underlying command manually (e.g., kubectl get pods) to verify access. Check the agent's required credentials in the docs.

Email us at ops@ternaryphysics.com.