Providers
Providers
Section titled “Providers”Ptah is multi-provider by design. You configure the AI services you want to use, Ptah routes chat traffic, sub-agent spawns, and tool calls across them, and a single unified cost ledger shows you what each turn cost — no matter which backend served it.
You bring your own keys. Ptah never proxies your traffic through a service we control. Your API keys stay on your machine.
Supported providers
Section titled “Supported providers”| Provider | Auth | Best for | Cost tracking |
|---|---|---|---|
| Claude | API key | Reasoning-heavy work, long contexts, prompt caching. | Anthropic published rates. |
| GitHub Copilot | OAuth | Code completion quality, bundled if you have Copilot. | $0 (your subscription). |
| OpenAI Codex | OAuth / API key | GPT-5 family, high-reasoning modes. | OpenAI published rates. |
| Ollama | Local / token | Offline work, privacy, local models. | $0 for local, Cloud uses Ollama rates. |
| LM Studio | Local (no key) | Local models over an OpenAI-compatible server. | $0 — runs on your hardware. |
| OpenRouter | API key | Access hundreds of models with one key; live pricing feed. | Live from OpenRouter registry. |
| Requesty | API key | Multi-model routing with no translation layer. | Requesty published rates. |
| Moonshot (Kimi) | API key | 256K context, agentic coding flagship (kimi-k2.7-code). | Moonshot published rates. |
| Z.AI (GLM) | API key | 1M context, open-source coding leader (glm-5.2). | Z.AI published rates. |
| Sakana (Fugu) | API key | Fugu and Fugu Ultra, 200K context. | Unpublished — reported as $0. |
| Ptah CLI | User-configured | Wrap any CLI-based agent. Highest priority in CLI detection. | Delegated to wrapped tool. |
Ollama Cloud is configured from the same panel as local Ollama. See Ollama.
You can also add a custom provider — any Anthropic-compatible or OpenAI-compatible endpoint you define yourself, with your own base URL, key, model list, and tier map:
ptah provider custom add my-endpoint --base-url https://... --key <key>And three web search providers for grounding:
- Tavily, Serper, Exa — API keys for web search tools.
How routing works
Section titled “How routing works”- The main chat agent uses the provider and model you pick with the in-chat model selector — or your default from
~/.ptah/settings.json. - Sub-agents spawned by Autopilot use the CLI detection priority (
ptah-cli > codex > copilot) unless the parent specifies a CLI explicitly. - Web search tools use whichever web search provider you have an API key for, with Tavily as the default.
- Embeddings / indexing tasks use a lightweight model from your active provider family.
See Switching for what happens when you change providers mid-conversation.
Where keys are stored
Section titled “Where keys are stored”Non-secret settings live in ~/.ptah/settings.json. Secrets live in secure storage, managed from the Providers settings page inside the app.
No provider gating
Section titled “No provider gating”Ptah is free and fully open source. Every provider integration — API keys and OAuth flows alike — is available to everyone. Bring your own key (or sign in via OAuth) and you’re set. No Ptah account, no license key, and no subscription is involved. See the individual provider pages for setup details.