Amp
Sourcegraph's spin-out agent with zero-markup model billing
About
Amp is the agentic coding tool that Sourcegraph spun out in December 2025 as an independent company founded by original Sourcegraph co-founders Quinn Slack and Beyang Liu — the same product path that absorbed the agentic ambitions of the now enterprise-only Cody. It runs from a CLI and integrates with VS Code, JetBrains, Neovim and Zed. The agent structure is unusually model-mixed: smart mode runs Claude Opus 4.7, deep pairs GPT-5.5 with extended thinking, rush runs GPT-5.5 without reasoning for cheap small tasks, and an Oracle tool gives the main agent an on-demand second opinion. Subagents handle search, library lookup and review on their own models. Threads are the unit of work and can be shared by URL and resumed later. A May 2026 rewrite of the CLI (shipped as Neo, GA later that month) added auto-compaction at 90% context, a TypeScript plugin API and remote thread control from ampcode.com — but also removed file rollback, user-invocable skills and, importantly, default permission prompts, so tools now run without approval unless you configure otherwise. Billing is the differentiator: there are no subscription tiers. You pay provider rates with zero markup, with a $5 minimum purchase. Amp Free was $10/day of credits (about $300/month) but stopped accepting new signups on 2026-02-10, and unused credits expire after a year of account inactivity. Execute mode and programmatic calls consume paid credits only. Enterprise is the same pass-through pricing plus a 50% markup and a $1,000 one-time onboarding fee, adding SSO, SCIM, zero data retention and an MCP allowlist. The catch: Amp publishes no per-token rates, so cost control depends on watching thread costs after the fact.
Key Features
- smart, deep and rush agent modes across frontier models
- Oracle second-opinion tool for hard decisions
- Parallel subagents for search, library lookup and review
- Shareable, persistent threads controllable from the web
- TypeScript plugin API and MCP support
- Zero-markup pass-through billing with a $5 minimum
Deals, Discounts & How to Save
Amp does not charge a markup, so your bill is whatever the models cost — which means the leverage is in model choice, not discounts. Route routine edits to rush mode, keep deep and smart for work that genuinely needs frontier reasoning, and remember that any free or promotional credits are separate from execute-mode usage, which always draws on paid credit.
Pros
- Zero markup on model usage — rare in the agentic market
- Mixed-model design puts cheap models on search and review subagents
- Threads are shareable and persistent, useful for team handoffs
- Spun out from Sourcegraph with backing from its original founders
Cons
- No published per-token rates, so cost is opaque until after the fact
- Free tier is closed to new signups
- May 2026 CLI rewrite removed file rollback and default permission prompts
- Enterprise adds a 50% markup plus a $1,000 onboarding fee
Pro Tips for Amp
Start with rush mode for small scoped edits — it is dramatically cheaper than frontier modes.
Watch the per-thread cost readout while a run is in progress; there is no upfront budget cap.
Keep long-running tasks in interactive mode: execute mode and programmatic calls only draw on paid credits.
Re-audit permissions after the May 2026 CLI rewrite — tools run without prompts by default now.
If you were an Amp Free user before February 2026, keep that account: the free daily credit pool is closed to new signups.
Alternatives to Amp
Cursor
FeaturedAI-first code editor built on VS Code
Cline
Autonomous coding agent in VS Code & JetBrains
Claude Code
FeaturedAnthropic's terminal-first AI coding agent
OpenCode
The most-starred open-source terminal coding agent
More in Cli Tool
Aider
AI pair programming in your terminal
Claude Code
FeaturedAnthropic's terminal-first AI coding agent
Gemini CLI
Google's free AI coding agent with 1M+ context
Codex CLI
OpenAI's Rust-powered terminal coding agent