FAQ

Common questions about Tahyi — what it is, how it works, and what to expect at launch.

General

How is Tahyi different from ChatGPT or a single coding agent for ops?

Tahyi is a coordinated swarm of domain specialists with safety gates and an audit log — not one generalist chat session. Each agent owns a narrow slice (deploy, monitor, observe, DBA) and hands off through explicit contracts in the coordination layer.

A single AI assistant has no persistent task model, no safety classification, and no cross-domain handoffs. Tahyi treats infrastructure operations as a system, not a conversation.

What does “swarm” mean in practice?

A swarm is multiple specialist agents plus the coordination layer that routes tasks, context, and safety gates between them. Specialists wake on triggers (schedules, alerts, human assignments), plan under policy, act on two-way doors autonomously, and gate one-way doors behind approval.

See Core Concepts for the full model.

Is Tahyi open source?

Tahyi is designed for self-hosted deployment with an MIT license planned for public release. Repositories are private during pre-release development — the board will approve open-source publication when ready. Track progress on TAH-74.

Safety and autonomy

Will Tahyi run destructive commands without asking?

No. One-way door actions — irreversible or high blast-radius operations — require explicit approval. Dry-run is the default on early builds. The agent plans, logs, and waits.

Two-way door actions (reversible within policy) may run autonomously.

See How it Works — Safety classification.

What counts as a one-way door?

Operations that cannot be cleanly reversed: destructive DDL, production data deletes, irreversible infra teardown, credential rotation that invalidates live sessions. The classification is explicit in every plan — not inferred after execution.

Can I force dry-run mode?

Yes. Dry-run is the default on pre-0.1 builds. You can also enable it per-environment or per-agent:

npx tahyi agent enable deployment-maintainer --environment staging --dry-run

Models and providers

Can I use my own LLM provider?

Yes. Tahyi is model-agnostic. Specialists connect to the provider you configure via environment credentials — Claude, OpenAI, Gemini, or open-source/local models.

Never commit API keys to the repo or agent prompts. Use .env or your secrets manager.

Does Tahyi train on my infrastructure data?

No. Tahyi runs on your infrastructure with your chosen model provider. Operational data stays in your audit log and environment — not in a vendor training pipeline.

Scope and roadmap

What infrastructure does Tahyi support at launch?

0.1 targets a single beachhead environment (documented stack). The four founding specialists — deployment, monitoring, observability, DBA — ship before breadth expands. See SCOPE for the full 1.0 boundary.

What is the release trajectory?

Version	Milestone
0.1	Single specialist runs real work end-to-end
0.2	Coordinated swarm — multiple specialists
0.3	Production-trustworthy safety model
1.0	GA — autonomous ops with gated one-way doors

Each 0.x delivers standalone value. See Core Concepts — Release trajectory.

What is out of scope for 1.0?

SecOps, FinOps, networking, and general-purpose agent building are out of scope unless the board moves the fence. Tahyi ships the ops swarm — not a toolkit for arbitrary swarms.

Getting started

How do I install Tahyi?

Clone the repo, install dependencies, configure your LLM provider, and start the control plane. Full steps in Quickstart.

How do I get help?

• Open a GitHub issue (when the repo is live)
• Follow TAH-74 for project updates
• Read Contributing for how to report bugs effectively