Elephant Goldfish Model

Open source alternative to Amazon Kiro

16.5K stars
Apache-2.0
Active this week
Last commit today

Last verified Nov 10, 2026

Build software specs with deep-context agents and stress-test them with fresh zero-context reviewers before writing a line of code.

Visit website GitHub16.5K

Read the full breakdownWho it is for, how it works, trade-offs, the quick start and the FAQ

who it's for

Who Elephant Goldfish Model is for#

Software engineers designing features with AI coding agents

EGM fits into the design phase of agent-assisted development. Before asking your coding agent to implement a feature, you run /egm-brainstorm and /egm-draft to produce a spec, then /egm-review to check it with fresh Goldfish subagents. The spec captures architectural decisions, edge cases, and trade-offs; the agent implements from that spec rather than from a rough prompt.

Skip if:

Your features are small enough that a one-line prompt to the agent produces the right result. EGM adds the most value at the boundary of non-trivial design decisions with multiple trade-offs.

Development teams adopting spec-driven workflows with AI agents

EGM gives a team a shared spec format (docs/designs/) and a repeatable review process. Engineers submit specs for Goldfish review before coding begins, and the drift report keeps specs accurate after implementation. Teams get a lightweight paper trail of design decisions without the overhead of a formal RFC process.

Skip if:

Your team already has a spec review process that works. EGM is most useful when the missing step is structured review before implementation, not when you already have that gate.

Engineering leads enforcing design review before implementation

The Goldfish review and /egm-drift-report skills give leads a way to verify that an AI-generated spec was stress-tested before code was written, and that the final implementation matches the approved design. The spec artifacts (docs/designs/, concept briefs) are version-controlled and reviewable in pull requests alongside the code.

Skip if:

Your review process happens at the code level and your team is not using AI coding agents for design. EGM's value is specific to the agent-assisted development flow.

the problem

The problem it solves#

AI coding agents are good at building what they are asked to build. The problem is that the person asking them has blind spots: unstated constraints, implicit assumptions, and design gaps that seem obvious in hindsight but are invisible at the time of writing. When an agent builds from a brief that contains those gaps, the resulting code reflects them too.

This surfaces late. A feature passes its own tests and fails at the integration boundary. An architectural decision turns out to conflict with an existing constraint nobody mentioned. An edge case that happens one percent of the time causes a data integrity issue in production. The earlier in the development cycle these failures appear, the cheaper they are to address. A bug in a spec costs minutes to fix; the same bug in shipped code costs hours or days.

how Elephant Goldfish Model solves it

How it solves it#

Elephant brainstorm and draft

The Elephant agent loads full project context including system history, architecture, and component boundaries, then explores the problem through open-ended questioning and structured brainstorming. It saves the agreed direction in a concept brief, then turns it into a living design doc under docs/designs/, section by section, pausing after each section for your feedback in chat.

Goldfish zero-context review

Fresh stateless subagents read only the draft spec and live source code, without any conversation history. A Goldfish swarm checks completeness, clarity, self-containment, and implementation readiness from independent, conflicting perspectives. The design is stress-tested before any code is written, exposing unstated assumptions the author and their context-loaded agent cannot see.

Drift report after implementation

After implementation, the /egm-drift-report skill compares the checked-in design documents with the current source code and reports where they diverge. This keeps specs accurate as the codebase evolves and surfaces when the implementation has drifted from the approved design.

Configurable review loop

The /egm-review-loop skill runs a configured group of Goldfish reviewers for a specific review task. You can use the stateless review pattern on non-software documents or with custom reviewers and standards beyond the built-in five-stage software spec review.

Bootstrap for existing codebases

The /egm-bootstrap skill creates or updates documentation for an existing codebase that lacks a spec or design docs. It is the entry point for teams adopting EGM mid-project, before they have a docs/designs/ directory to work from.

Multi-agent support

EGM installs as a plugin into Claude Code, Codex CLI, Antigravity CLI, and ChatGPT and Codex Desktop. Each agent gets the same set of skills; the install path differs per agent. Claude Code and Antigravity use a prefixed command format (/egm:egm-brainstorm); Codex uses a skill picker.

strengths · trade-offs

Strengths and trade-offs#

Strengths

Stateless review catches what context-loaded agents missThe core design principle is deliberate: Goldfish subagents start each review with no conversation history. This is not a limitation but the mechanism by which the tool surfaces blind spots. A reviewer who has seen the problem being discussed will unconsciously fill in gaps; a reviewer who sees only the document cannot. This is EGM's primary differentiator from single-agent spec tools.
Apache 2.0 license with no commercial restrictionsThe Apache 2.0 license allows unrestricted commercial and team use, modification, and redistribution. Teams can use EGM inside proprietary workflows, contribute back changes, or fork the plugin without opening their own code. There are no seat fees, no usage quotas, and no vendor lock-in.
Works inside your existing coding agentEGM installs into Claude Code, Codex CLI, Antigravity CLI, and the ChatGPT and Codex Desktop apps. Teams using multiple agents, or switching between them, can carry the same spec workflow across agent environments. The skills and reviewer prompts are part of the installed plugin, so they read consistently across each supported agent.
Built by Google engineers, 16,499 starsThe plugin was developed by and for Google engineers: the project website states that thousands of Google engineers use it. With 16,499 GitHub stars and 1,217 forks, EGM has seen broad adoption beyond Google. The project has an active contribution model with public issues, discussions, and contributing guidelines.

Trade-offs

-Requires a coding agent that can spawn fresh subagentsGoldfish reviews depend on the ability to create stateless subagent contexts. If your coding agent cannot create fresh subagents, EGM stops the review and reports the limitation. The README documents which agent environments support multi-agent execution (Claude Code, Codex, Antigravity) and which do not. Teams on agents without subagent support get brainstorm and draft but not Goldfish review.
-Review does not replace tests or human judgmentEGM's README states this explicitly: a Goldfish review checks the spec, not the code, and does not substitute for unit tests, integration tests, or human code review. It catches design gaps before implementation, not correctness bugs in the implementation itself. Teams expecting automated correctness verification will need to combine EGM with their existing test suite.
-63 open issues in active developmentWith 63 open issues, EGM is in an active development phase. Some agent-specific rough edges exist: the README documents known version-specific issues (for example, Antigravity CLI 1.4.8 does not complete prefixed command names, while 1.4.10 does). Teams on specific agent versions should check the issues list before expecting every feature to work as documented.

versus alternatives

Elephant Goldfish Model vs alternatives#

EGM vs Amazon Kiro

Both EGM and Amazon Kiro address spec-driven development before coding begins, but they take different forms. Kiro is a proprietary IDE with its own agent; EGM is an open-source plugin that installs into your existing coding agent.

Feature	Elephant Goldfish Model	Amazon Kiro
License	Apache 2.0	Proprietary
Form	Plugin for existing agents	Standalone IDE
Agents supported	Claude Code, Codex, Antigravity, ChatGPT	Kiro's own agent
Zero-context review	Yes (Goldfish subagents)	Not documented
Self-hosting	Plugin runs in your agent	No
Cost	Free (Apache 2.0)	AWS-priced

EGM is the better fit when you are already using Claude Code, Codex, or Antigravity and want to add structured spec review without switching tools. The Apache 2.0 license means no seat fees or vendor lock-in, and the stateless Goldfish review step is a structurally distinct mechanism for catching blind spots that single-agent spec generation does not include.

Amazon Kiro is worth considering if your team wants a tightly integrated AWS environment where requirements, design docs, and task sequences are managed inside a single proprietary IDE. Teams building on AWS who prefer a fully managed, end-to-end product experience over a plugin-based workflow may find Kiro's integration with AWS services a better fit for their setup.

install · quick start

Quick start#

bash

Install EGM into your AI coding agent using your agent's plugin system. The plugin source is available to download from GitHub Releases: https://github.com/google/elephant-goldfish-model/releases/latest

```bash
# Claude Code
claude plugin marketplace add google/elephant-goldfish-model
claude plugin install egm@egm

# Codex CLI
codex plugin marketplace add google/elephant-goldfish-model
codex plugin add egm@egm

# Antigravity CLI
agy plugin install https://github.com/google/elephant-goldfish-model
```

tech stack · detected from GitHub

What it's built on#

Languages: JavaScriptPython

frequently asked

FAQ#

Which coding agents does EGM support?

EGM supports Claude Code, Codex CLI, Antigravity CLI (command-line and app), ChatGPT Desktop, and Codex Desktop. Each has a different install path: Claude Code uses claude plugin marketplace add google/elephant-goldfish-model and claude plugin install egm@egm; Codex CLI uses codex plugin marketplace add google/elephant-goldfish-model and codex plugin add egm@egm; Antigravity CLI uses agy plugin install https://github.com/google/elephant-goldfish-model. ChatGPT and Codex Desktop apps use the published plugin listing in the desktop app. Goldfish review requires the agent to support fresh subagent creation; not all agents or versions support this.

What is the difference between the Elephant and the Goldfish?

The Elephant is a deep-context agent that holds your full project history, architecture, and component details. It runs the brainstorm and draft phases, exploring the problem and producing the spec with knowledge of everything that came before. The Goldfish is a set of stateless subagents that read only the draft spec and referenced source code, without any prior conversation. Because Goldfish reviewers start fresh, they catch assumptions the Elephant took for granted and gaps that only appear obvious from the outside.

Can I use EGM for documents other than software specs?

Yes, with the /egm-review-loop skill. The built-in /egm-review is designed for software design documents and runs a five-stage review. /egm-review-loop lets you configure custom reviewers and standards for other document types. /egm-bootstrap and /egm-drift-report are specific to software codebases; they compare documents against source code and generate documentation for existing codebases.

Is EGM free to use?

Yes. EGM is Apache 2.0 licensed, which means you can use it commercially, modify it, and redistribute it with no licensing fees. The plugin itself is free; you pay only for whatever AI coding agent you install it into. The Apache 2.0 license places no restrictions on commercial or team use and does not require you to open-source code you build with the tool.

Does EGM replace code review or automated tests?

No. EGM's README states this explicitly: a Goldfish review checks the design spec, not the code, and does not replace unit tests, integration tests, or human code review. Its purpose is to catch design gaps and unstated assumptions before implementation, so the bugs found are in the spec rather than in running code. After implementation, /egm-drift-report checks whether the code matches the approved spec, but it does not run the code or verify correctness.

also worth a look

Similar open-source tools#

BMAD-METHOD

Alternative toAmazon Kiro

Spec-driven AI development workflow that keeps context across sessions

55,088 GitHub stars
Licence MIT with trademark notice

OpenSpec

Alternative toAmazon Kiro

Spec-driven development for AI coding assistants

74,406 GitHub stars
Licence MIT

Hermes Agent

Featured

Alternative toGrok Bot

Self-hosted AI agent: persistent memory, multi-channel chat, model choice.

261,524 GitHub stars
Licence MIT

Claude Quickstarts

InAI & Machine Learning

MIT-licensed quickstarts for Claude agents, bots, and browser tasks.

20,022 GitHub stars
Licence MIT

Billion Context

Alternative toCompresr

Compression proxy for AI coding agents, enabling 5x token savings.

4,620 GitHub stars
Licence MIT with attribution requirement

Hive

Alternative toConductor

AI coding agent orchestrator for isolated parallel worktrees

475 GitHub stars
Licence MIT

by category

Open source tools in these categories#

617

12,942 GitHub stars
Licence AGPL-3.0
#Email

rakazo

Alternative toGrok Bot

AI teammates you own: your keys, your model, your machine.

7,478 GitHub stars
Licence Apache-2.0
#AI Agents

SnitchFeed

Live monitoring on LinkedIn, X, Reddit & more. Agentic social listening for startups.

Marketing & Growth

Advertise your product

The projects we add and every new guide. Free, unsubscribe any time.

Find open source alternatives to paid software

EGM vs Amazon Kiro

Feature	Elephant Goldfish Model	Amazon Kiro
License	Apache 2.0	Proprietary
Form	Plugin for existing agents	Standalone IDE
Agents supported	Claude Code, Codex, Antigravity, ChatGPT	Kiro's own agent
Zero-context review	Yes (Goldfish subagents)	Not documented
Self-hosting	Plugin runs in your agent	No
Cost	Free (Apache 2.0)	AWS-priced

Who Elephant Goldfish Model is for#

Software engineers designing features with AI coding agents

Skip if:

Development teams adopting spec-driven workflows with AI agents

Skip if:

Engineering leads enforcing design review before implementation

Skip if:

The problem it solves#

How it solves it#

Elephant brainstorm and draft

Goldfish zero-context review

Drift report after implementation

Configurable review loop

Bootstrap for existing codebases

Multi-agent support

Strengths and trade-offs#

Strengths

Trade-offs

Elephant Goldfish Model vs alternatives#

EGM vs Amazon Kiro

Quick start#

What it's built on#

FAQ#

Similar open-source tools#

Open source tools in these categories#

Developer Tools

AI & Machine Learning

LLMOps & AI Tooling

Find open source alternatives to paid software

Who Elephant Goldfish Model is for#

Software engineers designing features with AI coding agents

Skip if:

Development teams adopting spec-driven workflows with AI agents

Skip if:

Engineering leads enforcing design review before implementation

Skip if:

The problem it solves#

How it solves it#

Elephant brainstorm and draft

Goldfish zero-context review

Drift report after implementation

Configurable review loop

Bootstrap for existing codebases

Multi-agent support

Strengths and trade-offs#

Strengths

Trade-offs

Elephant Goldfish Model vs alternatives#

EGM vs Amazon Kiro

Quick start#

What it's built on#

FAQ#

Similar open-source tools#

Open source tools in these categories#

Developer Tools

AI & Machine Learning

LLMOps & AI Tooling

Find open source alternatives to paid software