Abacus AI CodeLLM

Hosted IDE from Abacus AI that runs Claude 4 Opus as its default backend agent (the Composer Agent) with multimodel routing, MCP support, built-in auth, and a polished web UI. Positioned as a Cursor alternative for users who want Opus quality without Cursor’s pricing or Claude Code’s CLI surface.

  • Site: codellm.abacus.ai
  • Vendor: Abacus AI
  • Pricing: ~$10/month (bundles CodeLLM + ChatLLM + Deep Agent + Computer Agent — verify current tier before citing)

Features

  • Composer Agent — Claude 4 Opus by default; Sonnet, Gemini 2.5 Pro, OpenAI models routable per task
  • Chat Mode — conversational pair programming, separate from inline edits
  • In-place inline edits + tab autocomplete
  • MCP support out of the box
  • Computer Agent — bundled browser-use equivalent
  • Built-in auth & UI — no separate Supabase wiring for simple apps

Why it matters

CodeLLM is a data point in two wiki threads:

  1. Planning-first AI coding — running Opus by default in a hosted IDE validates that the strongest models pair best with structured planning surfaces vs raw chat. Sibling claim to Claude Code Ultra Plan.
  2. SaaS death spiral — bundling four AI products at $10 is the kind of pricing that breaks per-seat models. Worth watching whether the bundle holds.

Caveats

  • Closed-source, hosted only
  • “Outperforms Cursor on SWE-bench” claim is from Abacus marketing — treat as positioning until third-party benchmark exists
  • Pricing tier may have shifted since the source video

Sources

See Also