Capabilities
The platform is agent-neutral: each coding agent is an adapter, and the scheduler picks a compatible agent, provider and model for every task. By default each agent runs on its own login.
| Capability | Aider | Claude Code | OpenAI Codex CLI | Gemini CLI | OpenCode |
|---|---|---|---|---|---|
| Verified CLI version | 0.86.2 | 2.1.281 | 0.157.1 | 0.61.0 | 1.18.32 |
| Structured events | — | ||||
| Session resume | — | — | — | ||
| MCP servers from the registry | — | — | — | — | |
| Several repositories | — | — | — | — | |
| Real task completed end to end | — | — | — | — | |
| Skills | |||||
| Add-on models (gateway) |
“Verified” means the adapter was checked against that CLI version's help output and real runs.
5 agents
Aider
A pair-programming agent for the terminal that works with most model providers. The adapter keeps Aider from committing, editing .gitignore or checking for updates.
Anthropic
Anthropic's coding agent, run headless with streamed JSON events. The most complete adapter — session resume, MCP servers, several repositories and structured rate-limit events.
Google's coding agent for the terminal, run headless with streamed JSON. The worker assigns each session an id; resume is not claimed because it could not be verified.
OpenAI
OpenAI's coding agent CLI, run with `codex exec --json` in a workspace-write sandbox. Supports session resume and structured JSONL events.
SST
A terminal coding agent that works with many model providers. Run with `opencode run --format json --auto` so permission prompts never block a headless run.