One harness. Every model. Total control.
The agentic harness for VS Code that unifies OpenAI, Claude, Copilot, Gemini, and 200+ OpenRouter models into one orchestration workflow. 13 AI agents, 14 MCP tools, 4 agent runtimes, project-adaptive setup wizard, and a Skill Plugins registry at skills.sh.
Switching between AI providers usually means switching tools, re-explaining context, and stitching outputs together by hand. Ptah unifies OpenAI, Claude, Copilot, Gemini, and 200+ OpenRouter models inside a single VS Code extension — with shared conversation context, real-time cost tracking, and a built-in setup wizard that scans your workspace, detects 13+ project types, and generates project-adaptive agents and CLAUDE.md rules before you write your first prompt.
The challenge
Building a single agentic harness that abstracts five incompatible provider APIs (OpenAI, Anthropic, Google, GitHub, OpenRouter) and four CLI agent runtimes — while preserving streaming, tool use, parallel dispatch, and a shared conversation thread across provider swaps.
The outcome
Engineers install one VS Code extension and instantly get OpenAI, Claude, Copilot, Gemini, and 200+ OpenRouter models with shared context, parallel CLI orchestration, skill packs from skills.sh, and a setup wizard that generates project-adaptive agents on day one — backed by docs at docs.ptah.live and a 100-day free trial.
Technical approach
- Unified provider harness — OpenAI, Claude, Copilot, Gemini, and 200+ OpenRouter models behind one switchable interface
- Shared conversation context — provider swaps preserve thread history; no re-priming when changing models
- Real-time cost & token dashboards — per-provider spend and usage tracked in the panel as you work
- Skill Plugins registry — install domain packs (Core, Angular, NX SaaS, React) one-click from skills.sh or the Skills Discovery panel
- Agent Orchestration — Gemini CLI, Codex SDK, Copilot SDK, and Ptah CLI run as parallel workers
- 6 MCP lifecycle tools — fire-and-check primitives for spawning, polling, reading, and stopping CLI agents
- Ptah CLI — Claude Agent SDK-backed runtime that connects 200+ models for headless workflows and A2A bridges
- Setup Wizard — 6-step automated workspace analysis, 13+ project type detection, LLM-generated rules and agents
- Built-in MCP server — Ptah exposes its own MCP endpoint so any MCP-aware client can use its 14 tools
- Secure per-provider key management — credentials scoped per provider, never shared across tiles
Results at a glance
- 13
- AI agents
- 200+
- LLM models
- 14
- MCP tools
- 4
- Agent runtimes
- 100 days
- Free trial
