Tools

Five tools. One window. Shared everything.

Two of these are built into OpenAxis: the chat and the orchestrator. Three are mature open-source tools, OpenCode, Open Design, and OpenWork, wrapped without touching a line of their source. Whichever you open, it uses the same API keys, the same theme, the same memory, and the same project context.

Built into OpenAxis

Chat

Talk to any model, without opening a second app.

A full chat built into the sidebar. Switch models mid-conversation, attach images and files, turn on web search when you need fresh facts, and pull a project's instructions into the thread. Every conversation is saved and backed up to disk.

  • Any model. Anthropic, OpenAI, OpenRouter, Google Gemini, local Ollama, or any OpenAI-compatible provider, picked from a menu at the bottom of the input.
  • Reasoning effort. For models that support it, set how hard the model thinks, remembered per model.
  • Attachments. Drop in images (PNG, JPEG, GIF, WebP) or text files.
  • Web search. Flip it on and the proxy enriches the context with live results.
  • Project context. Link a project to inject its custom instructions into the conversation.
  • History you keep. Saved automatically, backed up to disk, and pinnable to the top of the list.
Built into OpenAxis · the flagship

Orchestrator

One goal goes in. A team of agents runs with it. Out comes a folder of real files.

The orchestrator is an interactive assistant. You type what you need in plain language — "a pricing site with a checkout demo," "a Python library with tests," "a market research report" — and it plans the work as a visual DAG you can inspect. Each agent has a role: research, write content, design mockups, code, verify quality. They execute in dependency order and write real files to disk. A real deliverable, not a chat transcript.

Then a deterministic quality gate kicks in — plain code, no model. It statically checks the output for broken imports, missing files, dead asset links, unstyled pages, and APIs that don't line up between two agents' files. Defects force a fail no matter what the verifier model claims. Only the agent that owns the broken file gets relaunched.

Read the engine reference
  • Specialized agents. Research, work (content & design system), design (mockups), code, and a report-only verifier, each with its own rules.
  • It plans itself. A planning loop decomposes the goal, spawns sub-agents, and declares a file contract for each one.
  • Dependency graph. Agents declare what they depend on; upstream results feed downstream as context.
  • The model doesn't grade itself. Structural defects are caught by static analysis and force a fail, even if the verifier says it passed.
  • Self-correcting. When the gate fails, only the responsible agent is re-run, then the work is re-audited.
  • You can steer it. Give feedback on a finished run and it re-enters the corrective cycle with your notes.
OpenCode · by SST · wrapped unmodified

Code

A terminal-grade coding agent, in its own sandboxed view.

OpenCode is a serious code agent. It reads and writes files across your project and runs commands to get the job done. Inside OpenAxis it runs as a local server in a sandboxed view, themed to match the rest of the app, configured by the same proxy as everything else.

It's also the engine underneath. All three integrated tools drive OpenCode. That's why one config file can set them all up at once.

  • Reads & writes files. Works directly on your project on disk, not on pasted snippets.
  • Runs commands. Drives the toolchain to build, test, and verify, inside its own view.
  • Same keys, same models. Routed through the shared proxy, so it inherits whatever you configured once.
  • State is preserved. Switch to another tool and back; your session is right where you left it.
Open Design · by nexu-io · wrapped unmodified

Design

Generate and refine visual mockups, no terminal in sight.

Open Design produces interface mockups in HTML, CSS, and SVG and refines them across iterations. Normally it runs as a command-line daemon. OpenAxis launches it for you, captures it, and puts it behind a graphical view, so a designer who never opens a terminal can use it like any other app.

In an orchestration, the design agent receives the design system from the work agent and turns it into faithful mockups, which the code agent then reproduces.

  • HTML, CSS & SVG mockups. Real markup you can hand straight to a developer.
  • Iterative refinement. Reruns to fix thin or placeholder output until the mockup holds together.
  • No command line. The app starts the daemon and manages it. You just describe the screen.
  • Feeds the orchestrator. Mockups are written to disk and passed to the code agent to build for real.
OpenWork · by different-ai · wrapped unmodified

Work

Structured work and documents, tied to your projects.

OpenWork handles the structured side of building: content, documents, and organized work that isn't code or pixels. In OpenAxis it shares the same projects as everything else, so the instructions and context you set once follow you into Work and back out again.

In an orchestration it runs as the content engine, and it owns the design system: palette, typography, and the look the designer and coder then have to honor.

  • Documents & structured data. Writing and organized output in Markdown, JSON, and CSV.
  • Shared projects. The same project instructions used by Chat and the orchestrator apply here too.
  • Defines the design system. As an orchestrator agent, Work sets the colors and type the others must follow.
  • Native web search. OpenAxis adds search to Work through the proxy, not a built-in cloud key.
Built into OpenAxis · shared across tools

Memory

Stop re-explaining who you are and what you're building.

Memory is a profile plus a set of short, tagged facts *(under maintenance — coming soon)*. The profile describes you, your role, your stack, your preferences. The facts are reusable details. On each request, only the facts relevant to what you're asking get injected, inside a token budget, so the prompt stays lean instead of dragging your whole history along.

You stay in control. A global switch turns memory off entirely, and any single project can opt out on its own.

  • Profile. Free-form text about you, injected to personalize answers.
  • Tagged facts. Short, reusable notes; only the relevant ones enter a given request.
  • Auto-extraction. Optionally let a model pull new facts out of your conversations.
  • Off whenever you want. A global kill switch, plus per-project opt-out.
The glue

Projects tie it all together.

A project bundles instructions, a type, a preferred model, and context. It's how one setup reaches every tool.

Standalone projects

Tied to a single tool: Code, Design, or Work. Open the tool and the project's instructions are injected into the AI context. A standalone project can also become a building block inside an orchestration.

Orchestrator projects

These don't run alone, they coordinate several agents. They define the overall goal, the agents to chain together, and settings like auto-distribution, consistency checking, and retry-on-error.

Download it. Add your keys. Create.

Drag OpenAxis to your Applications folder, then add your API keys in the Config panel. Apple Silicon, macOS 14 and up.