Updated on
Amp: reviews and analysis
Sourcegraph's agent, no subscription: pay per use with a peculiar free tier.
Affiliate link · no extra cost to you
Amp comes from Sourcegraph, the code search house, and it shows in how it handles context on large repositories. Its economic model runs against the current: no subscription, you pay consumption, with a free tier sustained by discreet advertising. The agent is ambitious and picks models for you, which simplifies or annoys depending on your profile.
Best for: Teams with large repositories preferring consumption over subscriptions.
What the internet says
Its offer is aggressive: a free tier with daily credits that refill hourly and, on pay as you go, zero markup over the provider’s rate with a $5 minimum top-up. In exchange it asks for trust in a young team that has just spun out and that kills features with no backwards compatibility when it stops believing in them.
What the web repeats in favour
- A free tier with daily credits refilling hourly, without crippling the agent or requiring data sharing for training
- Zero markup over the provider rate on pay as you go, with a $5 minimum top-up and no subscription commitment
- It uses several models and picks by task, with four effort modes that put the cost in your hands
What the web repeats against
- The team states it keeps no backwards compatibility and removes features it stops using, so your workflow can break between versions
- The real price depends on model consumption rather than a flat fee, so monthly spend is not predictable up front
- The product has just spun out of its parent company and the enterprise plan requires an initial $1,000 purchase
Sweep sources: Docs · Press · Press · Communities
Pros / Cons
Pros
- Pay per consumption, no monthly fee
- Very good context handling on large repos
- Shared threads for teamwork
Cons
- Picks models for you, no manual choice
- Per-use cost is hard to predict at first
TLDR: Amp comes from Sourcegraph, the code search house, and it shows in how it handles large repositories. Its economic model runs against the current: no subscription, you pay consumption, with a free tier sustained by discreet advertising. The agent is ambitious and picks models for you, which simplifies or annoys depending on your profile. For teams with monorepos, its context is the argument.
What Amp is and how it works
Sourcegraph spent a decade building code search for the industry’s largest codebases, and Amp is that experience turned into an agent: a terminal and editor tool whose differentiator is understanding enormous repositories, where the problem is not writing code but finding what to touch among millions of lines.
Its two product decisions separate it from the herd. The economic one: no subscription, you pay credits per consumption, with a free tier sustained by discreet in-tool advertising. And the technical one: Amp picks the models for you per task, no selector, on the thesis that the tool knows better which engine suits each step.
It also adds shared threads: agent sessions are visible to the team, turning AI work into something reviewable and learnable between teammates.
What it is like day to day
Large repositories are where it earns its seat: Sourcegraph’s search inheritance translates into an agent that locates the relevant code in bases where others get lost or burn context exploring. For monorepos and projects with years of history, that difference is the product.
The absence of a model selector divides users predictably: for some, one less worry and good results. For others, a black box deciding costs for you. Pay-per-use has the same double edge: no fixed fee to amortize, but a bill that is hard to predict at first until you know your pattern.
Shared threads pay off in teams: watching how a teammate solved a similar task with the agent is free training, and that social layer of AI work is among the product’s most copyable ideas.
Pricing and plans
No subscription: pay-per-use credits, with an advertising-sustained free tier for light use. Real cost depends on your tasks’ volume and complexity, and the practical recommendation is a measured week of use before drawing budget conclusions.
Against the leaders’ $17-20 subscriptions, the math favors Amp on irregular use (you pay only when you work) and evens out or exceeds on intensive daily use.
Who it is for (and who it is not for)
Amp is for teams with large repositories where context is the bottleneck, and for anyone preferring to pay consumption over fixed fees. Shared threads add value in teams wanting to learn agent work together.
It is not for anyone who wants to pick the model at every step, nor for anyone needing predictable monthly cost, which is subscription territory.
Amp alternatives
Claude Code and Codex are the subscription leaders with more general judgment. Cursor gives the IDE-integrated route. For monorepos, Amp competes with a context advantage: for the rest, the choice returns to the usual suspects. They are in Amp alternatives and the ranking in the best AI coding tools.
Frequently Asked Questions
How does pay-per-use work?
You buy credits and each task consumes them according to its complexity and the models Amp decides to use. No monthly fee: quiet months cost little and intense ones show.
Why can I not pick the model?
It is a product decision: Amp assigns the engine per task, on the thesis that automatic selection outperforms manual. If fine model control is non-negotiable for you, this is not your tool.
What does the Sourcegraph inheritance contribute?
Context on large codebases: locating the relevant code in repositories of millions of lines is the problem Sourcegraph has spent a decade solving, and the agent inherits it.
What is the advertising-backed free tier?
A light usage level without paying, financed by discreet advertising inside the tool. Unusual in the category, and enough to evaluate the agent before buying credits.