What each product actually is
Amp — Sourcegraph multi-surface coding agent
Amp is Sourcegraph’s coding agent and development environment built for frontier agent work across surfaces. Per ampcode.com and the live Amp catalog page, the same agent and threads span terminal, web, macOS/iOS, and IDE-connected workflows. Product surfaces called out in vendor and catalog copy include remote Orbs (agents that keep working after you close the laptop), local Runners (amp --no-tui on servers/CI), Dial modes (low / medium / high / ultra), shared threads, MCP, plugins, and BYOK or linked ChatGPT (and other) subscriptions.
The published Amp review frames a terminal-native, code-intelligence-first story for large repositories — reason about the tree before editing — while current marketing emphasizes multi-surface continuity: review from phone or web, continue threads, and steer remote orbs alongside local work. Amp is not open source. Catalog marks it editorially recommended.
Live pages: Amp, Amp Review. Official: ampcode.com. Pricing: ampcode.com/pricing.
Goose — Agentic AI Foundation local agent (Desktop, CLI, API)
Goose is a general-purpose AI agent that runs on your machine. Per goose-docs.ai and the aaif-goose/goose project, it ships as a native desktop app (macOS, Linux, Windows), a full CLI for terminal workflows, and an API to embed it elsewhere. It is built in Rust and positioned for code, research, writing, automation, data analysis, and other local agent work — not only autocomplete-in-the-editor.
Goose connects to 15+ model providers (Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock, and more), including API keys or existing Claude / ChatGPT / Gemini subscriptions via ACP. Extensions use the Model Context Protocol (MCP); docs cite 70+ extensions. Recipes capture workflows as portable YAML. Goose is part of the Agentic AI Foundation (AAIF) at the Linux Foundation. Apache License 2.0 is confirmed on LICENSE (main).
Catalog places Goose in terminal CLI tools, AI CLI agents, and AI coding assistants. It is not marked recommended. Live pages: Goose, Goose Review. Official docs: goose-docs.ai.
Same intent, different packaging
Both answer “run an agent that edits, executes, and keeps going.” Shared intent is why this comparison exists. Amp leans commercial multi-surface continuity (threads across terminal/web/mobile, Orbs/Runners, Dial). Goose leans open local MCP-agent depth (Desktop + CLI + API, recipes, AAIF governance, BYOK/Ollama). Prefer fit over a one-point overall Score gap.
Current surfaces (per vendor pages)
Surfaces below come from live Amp and Goose product/docs pages and aicoolies catalog summaries as of 2026-09-25 (Europe/Istanbul). They support fit decisions; they are not a substitute for the locked Score table.
Amp
Multi-surface threads. Catalog and ampcode.com describe the same agent across terminal, web, macOS/iOS, and IDE-connected workflows — continue a thread from laptop, phone, or browser.
Orbs and Runners. Orbs are remote machines where agents run without supervision; threads get fresh orbs with code, plugins, and tools. Local path: run amp with local files, or amp sync to mirror orb changes. Headless / CI path: amp --no-tui on servers and runners, steered from web or phone. Vendor pricing also allows own runners for free on the Hobby tier.
Dial, MCP, plugins. Catalog copy documents Dial modes (low / medium / high / ultra), MCP, plugins, shared threads, and BYOK / linked subscription token sources with no Amp token fees on non-Enterprise tiers (confirm pricing).
Licensing / commercial shape. Proprietary freemium product — not OSS. Software access starts free (Hobby); optional Individual Megawatt / Gigawatt tiers, Teams workspaces, pay-as-you-go credits, and Enterprise.
Sources: ampcode.com · ampcode.com/pricing · aicoolies.com/tools/amp
Goose
Desktop + CLI + API. Install docs and home page cover Goose Desktop downloads (macOS / Linux / Windows) and CLI install via the stable download_cli.sh release script or Homebrew paths documented on goose-docs.ai. Desktop and CLI share provider, model, and extension configuration.
Providers and local models. Docs cover API keys, subscription bridges via ACP, OpenRouter and related routers, and local Ollama as part of the model-agnostic story.
MCP extensions, recipes, subagents, ACP. Docs and marketing describe 70+ MCP extensions, recipes as portable YAML workflows, subagents for parallel work, security controls (permissions / sandbox language), and ACP server / ACP provider paths for editor and subscription bridges.
Licensing. Apache License 2.0 — confirmed via LICENSE on aaif-goose/goose main. Software fee $0; model cost is BYOK, subscription bridge, or local inference.
Sources: goose-docs.ai · github.com/aaif-goose/goose · LICENSE · aicoolies.com/tools/goose
Scoreboard (from existing reviews)
Overall: Amp (score date 2026-03-25) 85; Goose (score date 2026-03-29) 84.
Speed: Amp (score date 2026-03-25) 87; Goose (score date 2026-03-29) 80.
Privacy: Amp (score date 2026-03-25) 83; Goose (score date 2026-03-29) 95.
Developer experience: Amp (score date 2026-03-25) 84; Goose (score date 2026-03-29) 78.
These figures come from the existing aicoolies reviews for each tool. Prefer fit and current product surfaces; use the table as supporting Score references with the dates above. This page does not claim a new head-to-head re-test.
Overall is a near tie (85 vs 84). Amp leads speed (87 vs 80) and edges DX (84 vs 78). Goose leads privacy (95 vs 83). That split matches the product-shape story: Amp’s multi-surface commercial agent with a speed niche on these reviews; Goose’s privacy niche for a local, extensible MCP agent.
Neither axis alone settles the buy. A team optimizing for Orbs continuity and commercial multi-surface threads may still pick Amp despite Goose’s privacy lead. A team optimizing for Apache-2.0 local control and MCP recipes may still pick Goose despite the one-point overall gap.
Workflow — multi-surface commercial vs local MCP agent
Daily loop
Choose Amp when the daily loop is a multi-surface coding agent: start in the terminal or web, hand work to Orbs that keep running after you close the laptop, review diffs and portals from phone or browser, and steer Runners in CI or servers with the same thread model. Speed 87 (score date 2026-03-25) supports that high-throughput agentic loop on the published review.
Choose Goose when the daily loop is a local agent you configure once and extend with MCP tools — Desktop for interactive sessions, CLI for terminal/CI-friendly runs, recipes for repeatable workflows, and providers that can stay on your keys or on Ollama. Privacy 95 (score date 2026-03-29) supports that local / BYOK story.
What “control” means on each product
Amp’s control story is operational continuity and commercial team surfaces: shared threads, Teams workspaces, Dial intensity, Orb access grants, and Enterprise audit/network controls when you need them. Goose’s control story is runtime ownership: Apache-2.0 source, local Desktop/CLI, MCP permission/sandbox language, and model path you choose (including fully local Ollama).
Licensing & billing
Amp — freemium commercial
Per ampcode.com/pricing and the live Amp catalog summary (as-of vendor pricing check reflected on aicoolies tool page):
Hobby (Free): all product features; pay-as-you-go orbs or free own runners; BYOK / linked ChatGPT and other subscriptions with no Amp token fees or limits on non-Enterprise tiers.
Individual: Megawatt $20/month and Gigawatt $200/month, each bundling included agent usage and orb minutes (vendor Individual cards describe ~45,000 minutes of orb time).
Teams: no extra workspace charge; members can sit on Hobby, Megawatt, or Gigawatt; shared threads, portals, plugins, pooled credits.
Enterprise: custom (SCIM, audit/network/MCP controls, pooled credits, priority support).
Cite catalog/vendor summaries; confirm plan details on the pricing pages linked in Sources.
Goose — $0 Apache-2.0 + BYOK / Ollama
Goose is Apache-2.0 open source at aaif-goose/goose. Software is $0; you pay model providers or run local models. AAIF / Linux Foundation posture is part of the governance story on the project home page.
Commercial fork
The clearest commercial contrast is Amp freemium proprietary product vs Goose $0 OSS software + model cost. That fork is independent of which Score dimension you prioritize. A near-tie overall (85 vs 84) means packaging and licensing fit matter more than chasing one Score point.
Who should pick which
Pick Amp if you want:
A Sourcegraph multi-surface coding agent across terminal, web, macOS/iOS, and IDE-connected threads
Orbs / Runners continuity (agents that keep going after you close the laptop; headless CI/server runners)
Dial modes, shared threads, MCP/plugins, and freemium → Teams → Enterprise packaging
The speed niche on these Scores (87, score date 2026-03-25) and the narrow overall lead (85 vs 84)
A recommended catalog default on the commercial multi-surface shelf
Pick Goose if you want:
An Agentic AI Foundation (AAIF) local agent with Desktop + CLI + API
Deep MCP extension ecosystem and recipe-based workflows
Model-agnostic BYOK or local Ollama with ACP bridges to subscriptions
The privacy niche on these Scores (95, score date 2026-03-29)
Apache-2.0 software at $0 with foundation-backed governance
Editorial winner: amp — fit-first multi-surface commercial agent default in a near-tie overall (85 vs 84). State Goose’s privacy niche explicitly so the page does not read like a landslide.
How this fits the Amp / Goose neighborhood
On aicoolies, Amp already sits next to Claude Code, Codex, Cline, Grok Build, and other commercial/OSS peers. Goose already sits next to OpenCode, Crush, Pi, Codex, Droid, and Claude Code on its alternatives set. Live Amp alternatives (2026-09-25) explicitly include Goose. Goose’s alternatives set does not yet list Amp.
Related live comparisons: amp-vs-cline, amp-vs-codex, amp-vs-claude-code, amp-vs-grok-build, goose-vs-opencode. Use those for editor HITL, commercial CLI, or Goose↔OpenCode OSS corners. Use this page when the fork is specifically Amp’s multi-surface commercial agent vs Goose’s open local MCP agent.
Hubs: /categories/ai-cli-agents, /categories/terminal-cli-tools, /categories/ai-coding-assistants.
Decision guide — five practical forks
1. Do you need agents that keep working after you close the laptop? If yes — remote Orbs, phone/web review, CI Runners — Amp is the natural fit. Goose can still run Desktop/CLI locally and in scripted paths; the remote multi-surface continuity thesis is Amp’s.
2. Do you need MCP-first extensibility and recipes as the core workflow? If yes — databases, browsers, internal APIs, portable YAML recipes — Goose is the natural fit. Amp documents MCP/plugins, but Goose’s documented MCP depth and recipes are the product emphasis.
3. Do you need proprietary freemium commercial packaging vs Apache-2.0 $0 software? Amp: Hobby free → Megawatt/Gigawatt → Teams → Enterprise. Goose: $0 software + BYOK/Ollama model spend. Confirm pricing and LICENSE pages linked in Sources for procurement.
4. Is privacy the primary Score niche? Goose leads privacy 95 vs Amp 83 (score dates 2026-03-29 vs 2026-03-25). If local/BYOK/Ollama posture is the buying constraint, Goose has the clearer Score niche — still choose on deployment reality, not Score alone.
5. Is speed the primary Score niche? Amp leads speed 87 vs 80. If interactive/agent throughput on the published review is the buying constraint, Amp has the clearer Score niche — still choose on product shape, not Score alone.
What “good” looks like for each buyer
Amp buyer profile
A strong Amp buyer wants one agent runtime across laptop, browser, and phone, with Orbs that continue unsupervised work and Runners for headless/CI paths. They may already live in Sourcegraph-line tooling or want commercial Teams/Enterprise controls. They shortlist Claude Code, Codex, Cline, Cursor, and Goose as peers — and Amp’s live alternatives page already lists Goose.
Goose buyer profile
A strong Goose buyer wants an agent that lives on their machine and grows with MCP extensions. They may use Desktop for interactive work and CLI for scripted or CI-adjacent runs. They care that providers are swappable and that recipes can capture repeatable workflows for a team. They often arrive from AAIF / Block lineage awareness or from Goose’s Discord/docs, and they shortlist OpenCode, Claude Code, Codex, Crush, and Pi as peers.
Shared intent, divergent constraints
Both answer “agent that edits and runs.” Shared intent is why the comparison exists. Divergent constraints — multi-surface commercial continuity vs open local MCP-agent platform — are why the winner is fit-first rather than “always pick the higher overall Score.”
Feature surfaces that matter in practice
Continuity vs locality
Amp’s documented Orbs / multi-device review path is a practical answer to “keep shipping while away from the desk.” Goose’s documented local Desktop/CLI + Ollama path is a practical answer to “keep inference and runtime on my machine.” Pick the continuity model that matches how your team works.
Tools and extensibility
Goose’s MCP extension catalog and recipe system are first-class. Amp’s strength is the multi-surface thread model plus Orbs/Runners/Dial and commercial team packaging, with MCP/plugins also documented on the catalog summary. For the latest extension and pricing lists, see the vendor pages in Sources.
Surfaces: phone/web vs Desktop/CLI
Amp: terminal + web + macOS/iOS + IDE-connected threads (catalog/site). Goose: Desktop + CLI + API (docs). Both cover more than a pure CLI, but the default “home surface” differs — Amp leads with multi-surface frontier agent continuity; Goose leads with native local agent.
Models
Amp documents BYOK and linked ChatGPT / other subscription token sources with no Amp token fees on non-Enterprise tiers (pricing FAQ language). Goose documents 15+ providers plus ACP subscription bridges and Ollama. Neither Score table is a model-quality league table.
Cost and procurement notes
Amp cost shape. Free Hobby entry; optional Megawatt $20/mo / Gigawatt $200/mo; Teams workspace at no extra charge; Enterprise custom. Budget orb minutes / credits and whatever model path you choose (BYOK or linked subs). Confirm ampcode.com/pricing if finance needs a one-pager.
Goose cost shape. Software $0 (Apache-2.0). Budget API keys, subscription bridges, or local GPU/Ollama ops. Free-credit offers sometimes mentioned in install docs are promotional — treat as time-bounded vendor offers, not permanent pricing.
Total cost of ownership is not overall Score. A near-tie overall (85 vs 84) means packaging and workflow fit matter more than chasing one Score point.
Privacy and data-path framing
On published aicoolies Scores, Goose leads privacy (95 vs Amp 83). Practical questions:
Where do prompts and repository context go by default for your chosen provider / Orb / runner path?
Can you keep inference local (Ollama) or on an approved gateway?
Can security review the agent runtime source (Goose Apache-2.0) vs accept commercial SaaS controls (Amp Enterprise audit/network/MCP controls)?
Do MCP extensions, plugins, or Orb grants introduce additional data paths?
Goose answers (2)–(3) with a local-agent + MCP story and scores 95 privacy (score date 2026-03-29). Amp answers with commercial multi-surface agent packaging and Enterprise controls when needed, scoring 83 privacy (score date 2026-03-25). Neither Score is a certificate that every deployment is air-gapped.
Speed and DX — what the leads mean
Amp leads speed (87 vs 80) and edges DX (84 vs 78) on the locked Score table. In buyer language: the Amp review’s agentic loop felt faster and slightly smoother for the scored workflow.
Goose’s lower DX Score is not a claim that Goose is unusable. It marks a different optimization: extensible local agent platform vs polished multi-surface commercial agent UX. If your rubric weights MCP/recipes/local governance above Orbs continuity, Goose can still win the purchase.
Overall 85 vs 84 is a near tie. Fit-first still applies.
Migration and coexistence patterns
From Goose to Amp. Teams sometimes want multi-surface Orbs continuity and commercial Teams packaging after prototyping with Goose’s MCP platform. Migration is workflow and billing, not a Score transfer.
From Amp to Goose. Teams sometimes need Apache-2.0 local control, richer MCP recipes, or AAIF governance after starting with Amp’s commercial agent. Again: workflow and licensing choice.
Coexistence. Some orgs keep Amp for product-engineering multi-surface sessions and Goose for automation/recipes/MCP-heavy local ops. That can be coherent. The editorial winner on this page remains amp for the default multi-surface commercial recommendation; coexistence is a team topology choice, not a winner flip.
When the near-tie scoreboard still helps
Overall 85 vs 84 is not a mandate. It still helps in three narrow ways:
Speed niche clarity. Amp 87 vs Goose 80 (score dates 2026-03-25 / 2026-03-29) matches the high-throughput commercial-agent emphasis without turning the page into a scare story about Goose.
Privacy niche clarity. Goose 95 vs Amp 83 matches the local-agent emphasis without pretending Amp cannot be used carefully with BYOK and Enterprise controls.
Anti-landslide framing. A one-point overall gap should never be sold as a wipeout. Fit-first amp is a default recommendation for multi-surface commercial buyers, not a claim that Goose buyers are wrong.
DX is a modest Amp edge (84 vs 78). Use it as supporting evidence, not as a fake landslide.
Practical evaluation checklist
Before you standardize on either tool, run the same short checklist on a real repository:
Install / account path. Can every engineer get onto Amp Hobby or Goose Desktop/CLI without a ticket maze?
Provider path. Does your approved model path work on day one — BYOK, linked subscription, or Ollama?
Surface path. Do you need phone/web Orb continuity (Amp strength) or local Desktop/CLI + recipes (Goose strength)?
Review path. Can security accept commercial SaaS + Enterprise controls (Amp) or prefer Apache-2.0 source review at aaif-goose/goose (Goose)?
Team path. Will the team live in multi-device threads, Desktop, terminal, or a mix? Match the surface to how people already work.
Use aicoolies Scores as a second opinion with the published dates — not as a substitute for that checklist.
FAQ-style clarifications
Is Amp open source? No. Amp is a commercial freemium product from Sourcegraph. Goose is Apache-2.0 open source under AAIF.
Is Goose only a coding autocomplete? No. Goose is documented as a general-purpose local agent (Desktop, CLI, API) with MCP extensions and recipes — code is a primary use, not the only one.
Does the overall Score mean Amp always wins? No. Amp is the fit-first winner in a near tie (85 vs 84). Goose remains the better pick when privacy / local MCP / Apache-2.0 fit matters more.
Can both use Claude or GPT models? Yes — both support BYOK / provider paths (Amp also documents linked ChatGPT and other subs; Goose documents ACP subscription bridges and Ollama). That does not turn either into Claude Code.
Are the Scores from the same day? No. Amp Score dated 2026-03-25; Goose Score dated 2026-03-29. Close in calendar time, still separate review artifacts — not a dual re-test for this page.
