Amp by Sourcegraph

Amp by Sourcegraph

Amp is a frontier AI coding agent by Sourcegraph that runs in the terminal with IDE integrations for VS Code, JetBrains, Neovim, and Zed. Multi-model: Claude Opus 4.7, GPT-5.4. Three modes: smart, rush, deep.

Free
Amp by Sourcegraph

Amp by Sourcegraph: A Cursor Alternative for Terminal-First Multi-Model Coding

Amp is a frontier AI coding agent developed by Sourcegraph that runs in the terminal with deep IDE integrations for VS Code, JetBrains, Neovim, and Zed. It supports multiple top frontier models simultaneously — including Claude Opus 4.7 and GPT-5.4 — and offers three distinct operating modes (smart, rush, deep) optimized for different coding scenarios. As a Cursor alternative, it targets developers who prefer terminal-first workflows, multi-model flexibility, and a pay-as-you-go pricing model with no platform markup.

Amp by Sourcegraph vs. Cursor: Quick Comparison

Amp by SourcegraphCursor
TypeTerminal-first CLI agent + IDE integrationsStandalone IDE (VS Code fork)
PricingFree tier + pay-as-you-go at LLM provider ratesFree / $20 / $40 per month
LLM choiceClaude Opus 4.7, GPT-5.4, and fast modelsBuilt-in models + own key
Offline / local modelsNoNo
Open sourcePartial (some tooling open)No
Codebase indexingNo (session-based context)Yes (automatic)
Multi-file editsYesYes

Key Strengths

  • Three Operating Modes — Smart, Rush, Deep: Amp offers distinct modes for different task types. Smart mode balances capability and cost for typical coding tasks. Rush mode prioritizes speed for fast iterations and quick fixes. Deep mode activates extended reasoning and more thorough analysis for complex architectural problems. This flexibility lets developers match the model's behavior to the actual complexity of each task, avoiding both over-engineering and under-investment in reasoning time.
  • Multi-Model Access with No Markup: Amp provides access to Claude Opus 4.7, GPT-5.4, and fast models in a single platform, charging at LLM provider rates with zero platform markup for individual users. This is a significant cost advantage over platforms that add percentage-based markups on top of API costs, particularly for heavy users leveraging frontier models daily.
  • Free Tier with No Ads: As of March 2026, Amp's free tier became completely ad-free, giving individual developers a genuinely useful free entry point that includes pay-as-you-go access to frontier models. This contrasts with Cursor's free tier, which caps usage more aggressively before requiring a paid subscription.
  • Thread Sharing for Collaboration: Amp supports saving and sharing entire coding sessions as threads, enabling developers to share not just code but the full context of an AI-assisted coding session — the conversation, the edits, and the reasoning chain — with teammates or the open community. This is particularly useful for onboarding, code reviews, and documenting complex debugging sessions.
  • Broad IDE Integration Coverage: Amp integrates with VS Code, JetBrains, Neovim, and Zed — covering a broader range of developer environments than most AI coding tools. Neovim and Zed support in particular addresses underserved communities of developers who prefer lightweight, high-performance editors over full IDE stacks.

Known Weaknesses

  • No Codebase Indexing: Unlike Cursor, which automatically indexes your entire codebase for semantic search, Amp relies on session-based context. This means developers must manually include relevant files or directories in each session, which can be cumbersome for large projects with many interdependent modules where implicit codebase knowledge is critical.
  • PAYG Cost Unpredictability: While the no-markup pricing is a strength, pay-as-you-go on frontier models like Claude Opus 4.7 can result in significant and unpredictable monthly bills for heavy users. Developers accustomed to flat-rate subscriptions like Cursor's $20/month plan may find budgeting harder with PAYG-only pricing at frontier model rates.
  • Terminal-First UX: Amp's terminal-first design, while preferred by many experienced developers, creates a steeper onboarding curve for developers who rely primarily on graphical IDE interfaces. The absence of a standalone GUI IDE means developers must be comfortable in terminal environments to get full value from the platform.
  • No Offline or Local Model Support: Amp does not support offline usage or local model providers like Ollama. All processing requires cloud connectivity, which can be limiting for developers in restricted network environments, security-sensitive codebases, or simply those who want model privacy guarantees that local execution provides.

Best For

Amp by Sourcegraph is best suited for developers who work primarily in terminal environments and want multi-model flexibility without platform markups — particularly those using Neovim or Zed who are underserved by most AI coding tools. It's an excellent fit for individual developers and small teams who want frontier model access (Claude Opus 4.7, GPT-5.4) on a pay-as-you-go basis without committing to a fixed monthly subscription, and who value the ability to share coding sessions as collaborative threads. The free tier makes it accessible for experimentation before committing to PAYG usage.

Pricing

  • Free: $0/month — free tier with PAYG access to frontier models, no ads (as of March 2026), thread sharing included
  • Pay-as-you-go: Charged at LLM provider rates with zero platform markup for individual users — exact cost depends on model choice and usage volume

Prices are subject to change. Check the official pricing page for current details.

Technical Details

  • Models supported: Claude Opus 4.7, GPT-5.4, and fast/lightweight models (multi-model platform)
  • Context window: Dependent on underlying model (Claude Opus 4.7: 200K tokens; GPT-5.4: varies)
  • IDE / platform: Terminal CLI (primary), VS Code integration, JetBrains integration, Neovim integration, Zed integration
  • Offline / local models: No
  • Codebase indexing: No — context is session-based; files/directories included manually per session
  • API access: Not publicly documented
  • Open source: Partial — some tooling available; core agent is proprietary

How It Compares to Cursor

Cursor is a standalone VS Code fork with a polished GUI, automatic codebase indexing, and a flat-rate subscription model. Amp takes the opposite approach: terminal-first, multi-model, pay-as-you-go, with no codebase indexing. The core tradeoff is convenience versus flexibility — Cursor gives you an out-of-the-box AI IDE experience with deep codebase awareness; Amp gives you frontier model access across more IDEs at provider-rate pricing, but requires more manual context management. For developers who are already power users of the terminal and want to pick their model for each task, Amp offers more control at potentially lower cost.

Conclusion

Amp by Sourcegraph is the right Cursor alternative for terminal-native developers who want multi-model access to frontier AI coding capabilities without platform markups, particularly those using Neovim or Zed. If you value model choice, PAYG pricing transparency, and collaborative session sharing over the polished GUI and automatic codebase indexing that Cursor provides, Amp is one of the most capable and cost-transparent options in the current AI coding agent landscape.

Sources

FAQ

Is Amp by Sourcegraph free?

Yes. Amp offers a free tier with pay-as-you-go access to frontier models at LLM provider rates with no platform markup. As of March 2026, the free tier is also completely ad-free.

Does Amp work with VS Code?

Yes. Amp integrates with VS Code as well as JetBrains, Neovim, and Zed, in addition to its primary terminal CLI interface.

How does Amp compare to Cursor?

Cursor is a VS Code fork with a full GUI, automatic codebase indexing, and flat-rate subscriptions. Amp is terminal-first with no codebase indexing but offers multi-model access (Claude Opus 4.7, GPT-5.4) at provider rates with no markup, plus broader IDE coverage including Neovim and Zed.

What are Amp's three operating modes?

Amp offers Smart (balanced capability/cost for everyday tasks), Rush (speed-optimized for fast iterations), and Deep (extended reasoning for complex architectural problems) modes, letting developers tune the agent's behavior to each task's complexity.

Reviews

No reviews yet

Similar alternatives in category