One command.Every agent.
Orchestrate Claude Code, Gemini CLI, and Codex CLI in parallel from a single desktop app. Ship 10x faster with multi-agent AI development.
Features
Everything you need to ship with AI agents
01
AI CLI Multi-Provider
Claude Code, OpenAI Codex, Gemini, and local open-source models via Ollama, vLLM, LM Studio, or any OpenAI-compatible endpoint — all in a unified chat. Per-session model memory, real-time streaming, extended thinking. Works with your subscription, zero API credits.
02
Parallel Sessions
Fan one prompt out to Claude Code, Codex, Antigravity, and your local open-source models at once. Each agent attacks the same task autonomously while a live council card streams their progress — then a synthesizer reads every draft, benchmarks them 0–10, and reconciles the strongest ideas into one definitive plan handed to native Plan Mode for your approval. Launch it from the composer's Parallel pill, @oxima-parallel-all, or /code-review.
03
Design Mode
Build interactive webapps inside OXIMA. Click any element in the live preview, write a feedback comment devtools-style, and the agent applies the change. Auto-paused dev server during edits, automatic git checkpoints per turn, viewport toolbar, export as Markdown blueprint or save to disk.
04
Live Share & Hosting
Ship the project, not the source. Build a production output of your app and publish it to share.oxima.ai in one click — the client only ever sees the resulting site, never your code. Manage expiring share links from a single panel: extend by +7 days or revoke instantly. From the same codebase you can host on Cloudflare or scale all the way to a production SaaS, never rewritten.
05
Agent System (@mentions)
Domain experts you activate inline with @mention, universal across every provider. 9 bundled agents — @typescript, @react, @nextjs, @vue, @svelte, @tailwind, @supabase, @prisma, @tauri — each with its own knowledge sources and AI-assisted updates. Activation is session-scoped and cumulative, with a learning loop that distills reusable rules from your chats.
06
Automation & Scheduling
Schedule recurring AI tasks — intervals or specific 24h times, weekday filters, target an existing session or spawn a new one. 18 ready-made templates, full execution history, dual-layer concurrency guard, crash recovery on restart.
07
Service & MCP Catalog
Plug external services into the agent two ways, no aggregator hub. CLI services: one-click install + auth for 14 service CLIs — Supabase, Vercel, Firebase, Stripe, GitHub, Cloudflare and more — surfaced into the agent's system prompt so it knows what to hand off to. MCP servers: a curated catalog (plus any custom server) with lazy npx/uvx install, a read-only tool inspector, and cross-CLI token sharing.
08
Prompt Library & Custom Commands
36+ prompt templates and 37+ slash commands with fuzzy multi-tier search. Define your own custom prompts and trigger them with /. 34 scaffolding templates across 12 frameworks (Next.js, Vue, SvelteKit, Expo, Flutter). Extensible with the Agent Skills Open Standard.
09
Telegram Bridge
Drive OXIMA from your phone. One forum topic per chat session with full history, voice input transcribed via Whisper/Gemini, and photo attachments. Answer permission requests, AskUserQuestion, and plan reviews from inline keyboards; switch model with /model and browse your prompt library with /prompts.
10
Native Performance — Rust + Tauri
Tauri 2 desktop app on a Rust backend, not Electron. Concurrent streaming across providers without blocking, sub-200ms session cancel, atomic state persistence, process-group cleanup so child processes never leak. Native macOS feel with private API + overlay titlebar. The whole app ships as an ~18 MB installer — a fraction of the hundreds of megabytes a comparable Electron build would weigh.
And more
Four Interaction Modes
Ask, Auto Edit, Full Auto, Plan Mode — granular control over what the AI can touch, persistent per project
Built-in Terminal
Native PTY (xterm.js + portable-pty), zsh and bash login shells, real-time streaming, dynamic resize on pane changes
Git Integration
Branch selector, commit and push from the app, auto-refresh staged/unstaged diff, git status injected into prompts
Theme System
24 preset themes, custom color picker (base + accent), sidebar transparency 0-100%, zero re-render on theme switch
Web Capabilities
WebFetch and WebSearch pre-approved across Claude, Codex, and Gemini for live information access
Voice Input
Record audio directly from the input field, transcribe via Whisper — no API key required
Advanced Input
@ file mentions, / slash commands, IDE context toggle (git status + diff stats), thinking toggle, model selector pill
Permissions System
Hook system intercepts PreToolUse and PermissionRequest with smart timeouts, Active and Spectator modes
Multi-Project Workspace
Multiple projects open in one app, isolated state and chat per project, instant switching with full state restore
Skills System
Extend the AI with custom cross-CLI skills (Agent Skills Open Standard) — auto-discovery, Skill Creator, Installer with cross-CLI sync, Skills Browser; skills show up as / commands in chat
Native MCP Server
9 exposed tools (forge_get_tasks, forge_get_current_task, forge_remember, forge_recall, etc.), HTTP hook server for real-time events
Usage & Cost Tracking
Per-session quota tracking with token and cost breakdown per provider and model — real data for Claude and Codex, request-count for Antigravity
Context Panel
Chat sidebar with Progress Card stepper and Artifacts Card showing AI-touched files with click-to-preview
Popout Chat
Detach the chat into a separate window with full functionality — ideal for multi-monitor workflows
Pricing
Simple, transparent pricing
Pricing will be finalized closer to launch. Join the waitlist for early-bird rates.
Open Source
Pro
Team
Changelog
What's new in OXIMA
Parallel Sessions & Design Mode redesign
- Parallel Sessions: fan one prompt out to Claude, Codex, and Gemini at once — an orchestrator benchmarks every result (0–10) and synthesizes a single plan you review
- Figma-style Layers Panel with two-way selection and a DOM ⇄ source-file view
- Per-element inspector — text, color, radius, shadow, border, opacity, add classes, live text editing — committing to the exact source line
- Motion Picker: visual animation presets compiled into framer-motion / GSAP
- Draggable canvas dock with a bottom-sheet inspector and git-backed snapshots (apply-to-main)
Agents, Telegram & Service Catalog
- Agent System: 9 bundled domain-expert agents via @mention, with per-agent knowledge + learning loop
- Telegram Bridge: run sessions from your phone — voice/photo input, inline-keyboard prompts
- Service & MCP Catalog: one-click install + auth for 14 cloud services, curated MCP servers
- OpenRouter + more OpenAI-compatible connectors, Design Share ephemeral links, Claude Opus 4.8
- New OXIMA theme — 24 presets total
Design Mode & Local Models
- Design Mode: build interactive webapps in a Vite+React+Tailwind sandbox, edit via devtools-style comments, automatic git checkpoints
- Open-source local models: Ollama (Qwen 3, Gemma 4, DeepSeek R1, GLM-4), vLLM, LM Studio, plus OpenAI-compatible cloud (Groq, Together.ai, Fireworks.ai)
- Project File Explorer with 36+ language icons and attach-to-chat
- Integrated git controls (branch selector, commit, push with auto upstream, diff viewer)
- Universal preview updated to 8 formats (code, HTML, Markdown, images, SVG, PDF, JSX, text)