5 providers 9 models v1.0.10

THE RIGHT MODEL
FOR EVERY TASK.

The first CLI tool with true multi-model orchestration. Assign Claude for planning, GPT for coding, Gemini for reviews — all coordinated in split-panel terminal panes.

src/core/model-router.ts esc when done

                            12
                            export const resolve = ({ role, ctx }) => {
                        
                            13
                              const config = loadRoleConfig(role);
                        
                            14
                              const primary = config.primary ?? defaults[role];
                        
                            15
                              if (isAvailable(primary)) return primary;
                        
                            16
                              return config.fallback.find(isAvailable) ?? ctx.default;
                        
                            17
                              return <ModelStream provider={primary} />;
                        
                            18
                            }

01 [x]

Multi-Model Orchestration

Claude Opus for architecture, GPT-5.2 for implementation, Gemini 2.5 Pro for code review — all in one session with real-time streaming.

02 [x]

Split-Panel Mode

Each agent gets its own iTerm2 or tmux pane. Hub-and-spoke coordination with shared board directory and real-time IPC.

03 [x]

Cross-Model Agent Teams

LLM designs the team from natural language. No hardcoded agents — roles, models, and tasks are all dynamically decided.

04 [x]

Role-Based Routing

4-step model resolution: user override, role config, fallback chain, system default. Planning, coding, review, testing — each gets the best model.

05 [x]

Native Account Login

Login to Claude, Codex, Gemini, and Kimi via their native browser auth flows. Use your existing Pro/Plus subscription — no separate API billing.

06 [x]

Enterprise Security

OS keychain credential storage, AES-256-GCM encrypted fallback, HMAC-SHA256 IPC authentication, sandboxed shell execution.

> agent teams

LLM-Designed Agent Teams

Describe what you need in plain English. The LLM analyzes your prompt and designs the optimal team — agent count, roles, models, and task prompts.

Create a team to refactor the auth module

I need agents to review this PR from different angles

Spawn a team to build the payment feature

Split-Panel Output

> Analyzing team request... Done

> Designed 3-agent team:

AuthArchitect (claude-opus-4-6) — planning

Refactorer (claude-sonnet-4-6) — coding

SecurityReviewer (gpt-5.2) — review

> Split-panel active ✓

/team stop to exit

THE RIGHT MODEL FOR EVERY TASK.