Choose a Session Backend

Yeaft puts three AI backends side-by-side in the same Web UI — you pick one per task. They're not mutually exclusive: a single Agent can run Claude Code sessions, Copilot sessions and Yeaft sessions in parallel without interfering with each other.

Backend	Good for	Not ideal for
Claude Code Chat	Long project collaboration, deep code understanding, requires /skills and MCP tools	No Claude Code CLI installed; budget-tight workflows
Copilot Mode	Existing GitHub Copilot subscription, want to swap between GPT-5 / Gemini / Claude on demand, standardized ACP protocol	Need /compact; depend on Claude-only skill system
Yeaft Sessions	Multi-VP parallel discussion, cross-session persistent memory, custom VPs (persona + model + tools)	Single-thread 1:1 chat (Yeaft sessions are multi-VP today)

Core differences

Claude Code Chat (1:1 chat backed by Claude Code CLI)

• Each session = one Claude Code CLI subprocess
• Full Claude Code stack: skills, MCP, subagents, /compact, /clear, /btw
• Tool calls go through Claude Code's own stream-json protocol
• Session history lives in ~/.claude/projects/ and can be resumed

Copilot Mode (1:1 chat backed by GitHub Copilot CLI)

• Each session = one copilot --acp subprocess, talks ACP (Agent Client Protocol)
• Multiple models available (Claude Sonnet 4.x / Claude Opus 4.x / GPT-5.x / Gemini 2.5 Pro etc.), gated by your Copilot subscription
• Tool permissions confirmed per call via ask-user popup (you can toggle "Allow all tools" to skip)
• Session history lives in ~/.copilot/session-store.db and can be resumed

Yeaft Sessions (multi-VP collaboration on the in-house engine)

• No external CLI dependency — Yeaft ships its own query loop, memory and tools
• A session hosts multiple VPs (Virtual Persons) — each VP has its own persona, model and toolset
• A single user message fan-outs in parallel to multiple VPs, each replies independently
• H2-AMS persistent memory — keeps vp / group / user / feature / global scopes across sessions
• Multi-provider LLM: configure OpenAI / Anthropic / GitHub Copilot in any combination via ~/.yeaft/config.json

How to pick in the UI

Chat / Crew Mode (Claude Code or Copilot)

Sidebar + opens the session config modal:

1. Pick Agent (machine)
2. Pick Provider: Claude Code or Copilot
3. Pick Working directory
4. If Copilot, a model picker and Allow all tools checkbox appear

Yeaft Sessions

Sidebar tab bar at the top → switch to Yeaft, then use + to create a session:

1. Enter session name
2. Pick the VPs you want in this session (VPs are reusable; build them in the VP Library)
3. Send a user message; the VPs reply in parallel (use @VPname to address a subset)

Which one should I use?

• Write code daily + already using Claude Code → Claude Code Chat
• Want it cheaper / want GPT-5 / company has Copilot Enterprise → Copilot Mode
• Want "PM + Dev + Reviewer" to discuss a feature with you simultaneously → Crew Mode (on top of Claude Code)
• Want multiple VPs to remember you long-term / continue memory across tasks / mix OpenAI + Anthropic freely → Yeaft Sessions

Next:

• Claude Code Chat
• Copilot Mode
• Yeaft Sessions