PhengosA native AI workspace that runs against your own provider keys or entirely offline.

Two ways to run ​

Phengos supports two execution paths in a single native application:

• Bring your own key for any of the eight online providers — Anthropic, OpenAI, Mistral, OpenRouter, DeepSeek, Zhipu (GLM), Qwen, or Moonshot (Kimi) — and talk straight to the provider, with no accounts, no middleman, and no markup.
• Run entirely offline through Ollama: pick Ollama as the provider, choose one of your installed local models, and generate. No API key, no per-token cost, and nothing leaves your machine.

The provider is selected per run. The workflow, saved history, and verification are identical in both cases; only the location of the model changes.

Offline operation ​

Confidentiality requirements, regulatory exposure, cost control, and availability constraints all rule out sending prompts and source material to a hosted model. Phengos was designed local-first, and the offline path is a fully supported provider running the same workflow as any other.

• Nothing leaves your machine. Local Ollama runs make no network request at all — no prompt, response, or file is transmitted. The session cost reads local / $0.0000 because there is nothing to bill.
• No key, account, or vendor dependency. Pull an open model with ollama pull, select Ollama in Phengos, and generate. The model and the hardware are yours.
• Provable output, even with no internet. In Pro, each local run is sealed with the same SHA-256 integrity record as a cloud run (see below) — so an air-gapped, no-cost model still produces work you can verify later.

Availability. Ollama is a desktop provider, available in every edition on Linux and Windows. Image and file inputs are not sent to Ollama in the current version — local runs are text-only; use a cloud provider when you need to send images as context.

Scope ​

Conventional AI clients present a chat interface. Results are not retained in a form you can audit, execution requires a hosted service, and the application itself carries a web dependency stack that must be kept patched.

Phengos addresses all three. It is native desktop software licensed perpetually, it runs online or fully offline, and it stores each run with its inputs, outputs, and generated files in a project directory on local disk, available for reproduction or verification later.

Reproduce and verify any run ​

Open a project months later and inspect exactly how a result was produced:

• the prompt you sent
• the exact system prompt the model received
• the model and settings used — cloud or local
• the response
• every generated file
• (in Pro) a SHA-256 integrity record and chain hash

Each record is searchable, reusable, and, in Pro, re-verifiable to confirm that nothing was altered after generation. This applies equally to cloud runs and offline runs.

Interface ​

Context files, prompt composition, provider selection, and saved output are presented in a single native window.

Projects and sessions ​

Every run is saved as a self-contained folder, organized into projects and sessions that can be reopened, searched, and reused.

Output to project directories ​

Generated code and content are written directly to the project output directory and tracked against the run that produced them.

Dependency profile ​

Phengos is written in native C++ on the Hyperoche Engine, our in-house film-grade real-time rendering engine. That engine is the only framework it bundles; everything else relies on the libraries your operating system already maintains.

• No Electron, no embedded browser, no Node runtime.
• Nothing is downloaded and run at launch — no curl | sh bootstrap, no plugin downloads, no auto-updating script engine. What you installed is what runs.
• The riskiest pieces stay with your OS. TLS and system crypto are patched by your distro or Windows Update — immediately, with no Phengos release required.
• No bundled dependency tree. Phengos does not carry the update churn or supply-chain surface associated with bundled package stacks.

Reusable workflows and profiles ​

Standing instructions are separated into two reusable, composable definitions rather than repeated in each prompt:

• Workflow — who the model is and what it's doing. For example, "a senior C++ engineer debugging this codebase," along with recommended settings for that kind of work.
• Profile — what it's working on and the conventions to follow. For example, "C++ headers and source,"

The two are combined per project, so one workflow can target different profiles without modification. Both are optional. See Workflows & Profiles.

Editions at a glance ​

One license. Every supported platform. Your keys — or no key at all. Your data. Pro is a one-time perpetual license — you own this version forever, with no subscription and no per-device charge.

Download Community free · Buy Pro on itch.io · Compare editions

Community edition ​

The Community edition is permanently free and provides the complete Phengos workspace rather than a limited trial. Bring your own key for any supported provider, or run local models through Ollama fully offline at no cost, with unlimited generations. Provider access is identical across all three editions. Your keys stay in your OS keyring and your work stays on your machine, exactly like Pro. The only limits are practical: up to five reference and five source files per run, with persistent history, sessions, and integrity verification reserved for Pro.

When you need integrity verification across both cloud and local runs, persistent reproducible history, and multi-turn chat, Pro is $49 as an introductory price — a one-time perpetual license (regular price $79).

Enterprise covers editable provider endpoints, negotiated support, and custom terms. See the editions page for details.