BETA · GAMMA-Ray is published for review. The rules, the mapping and every rating here are experimental and will change. How a rating is made · what exists · send a correction
GAMMA-Rayv0.0.155 · beta

Replit

GAMMA-Ray report · Spectrum 0.2 beta · Public tier · read 2026-11-10 · not reviewed by the company yet

Replit runs a cloud platform where people describe software in plain language and an AI agent writes, runs, tests and publishes it.

1solution carded
of 13 found
28sources captured
and hashed
76provisions lit
in 8 of 8 instruments
6examples of the agent
at work, with sources

Built only from Replit's own public pages (documentation, product, security, subprocessors, customers, newsroom, about) and, for founding facts only, Wikipedia. Replit has not seen this report. Nothing was tested and no credentials were used. If you are Replit and something here is wrong or missing, the correction will be made and dated, and you can claim the card to add evidence.

The agentic solutions

One card per solution. Each level is computed from the evidence by a published rule; a dimension with nothing public reads "not yet evidenced", which is not a no. A card rates a product as documented, not anyone's deployment of it.

Replit Agent (Agent 4)

Builds apps, designs, slides, data visualisations and files from plain language; writes code, sets up infrastructure, tests and publishes. Modes: Free, Power, Max, Auto (Enterprise: Auto).

Replit Agenta product, as documented
SPECTRUM 0.2 BETA · Public tier · 2026-11-10
A4AutonomyOrchestrateswith options on · as shipped A3
π Moderate · S1·R5·A1 ×3
R2ReachIntegratedwith options on · as shipped R1
π Moderate · S1·R5·A1 ×3
C2ControlConfigurablewith options on · as shipped C1
π Moderate · S1·R5·A1 ×3
O2OpennessDocumented
π Moderate · S1·R5·A1 ×3
DraftStage: Scaling (10 of 16)In production

Standout. Autonomy 4 (Orchestrates): "Read and write to external services like BigQuery, Linear, Slack, Notion"

Next unlock. Reach 4 (Ecosystem): public evidence that agents operate across organisational boundaries or disparate stacks would get there.

Trust. FACT-Ray: 28 facts: 28 π Moderate · found by an agent, captured and hashed, checked by 2 other agents (same model family, shared repository); no human has signed it.

Standards constellation. 76 provisions lit across 8 of 8 instruments.

SalesforceEAAMMOWASPMCPAAGMMMicrosoftNISTAI Act

Evidence: 18 sources, captured and hashed · the card's page has the vault's read key · Not reviewed by its vendor yet · correct or claim this card

The full card, every answer and its evidence · the governance layer

General Agent

Agent without a pre-selected artifact type: research, files, executes code, reads and writes connected services. No card yet (GA): this solution has not been assessed.

Plan mode

Agent plans without changing code or data; the user approves the task plan before building. No card yet (GA): this solution has not been assessed.

Task system / background tasks and worker chats

Agent tasks run in isolated copies of the project; Ready tasks are reviewed before being applied to main; a main chat can hand independent tasks to parallel worker chats. No card yet (GA): this solution has not been assessed.

App Testing

Agent tests the app in a real browser, records a replay and fixes issues; user can take over for logins. Toggle in Power/Max. No card yet (GA (Full Stack JavaScript and Streamlit web apps only)): this solution has not been assessed.

Routines

Scheduled recurring runs from a chat that use connected tools, create tasks and send messages; Power or Max only. No card yet (GA): this solution has not been assessed.

Connectors, custom connectors and MCP servers in Agent

First-party connectors (BigQuery, Linear, Slack, Notion, etc.), curated third-party MCP servers, custom MCP servers and custom API connectors that Agent can read and write through. No card yet (GA (Databricks/Snowflake Enterprise; Microsoft Fabric preview)): this solution has not been assessed.

Replit MCP Server (Replit in ChatGPT, Claude, Codex, any MCP client)

Exposes Replit to other AI agents over MCP (Streamable HTTP, OAuth with protected-resource discovery) to create, inspect, update and publish Replit Apps. No card yet (GA): this solution has not been assessed.

Replit on Slack

Agent answers and works on projects from Slack mentions and DMs, using the linked user account permissions. No card yet (GA): this solution has not been assessed.

Agent Skills, Skills directory and Agent Customization

SKILL.md skills (project or workspace), Replit and partner skills, workspace custom instructions, replit.md. No card yet (GA): this solution has not been assessed.

Security Agent (Agent security scans) and Auto-Protect

Agent-run source code security review with Semgrep and HoundDog.ai plus model analysis; accepted findings become Agent fix tasks. Auto-Protect creates Agent tasks for new dependency CVEs. No card yet (GA (paid plans)): this solution has not been assessed.

Monitoring Agent

Investigates production alerts; with Auto-fix attempts fixes; with Auto-republish (experimental) publishes them. No card yet (GA; Auto-republish experimental): this solution has not been assessed.

Performance Agent, SEO Agent

Listed in the docs feature index; pages not read for this dossier. No card yet (listed in docs index (not read)): this solution has not been assessed.

Our reading

The shape. Replit Agent reads A4 · R2 · C2 · O2 (Scaling). Its power is documented in plain terms: Agent builds, runs, tests and publishes on its own, and with connectors it reads and writes to outside services such as BigQuery, Linear and Slack. Its controls are settings a person can see: background work waits for review by default, there is a button to halt it, and Enterprise admins choose the scopes builders can request.

Openness. Replit documents Agent's limits and failure cases, the data it uses and its model providers. Under Spectrum 0.2 that meets Disclosed and Documented; the next rung asks for published evaluation results.

Where the evidence is strongest. The autonomy story is backed by named customers: Forbes, in an article on Replit's newsroom, reports that Zillow's employees made more than 7,000 apps with Replit over a year on about 600 seats, and Plaid built an SLA dashboard with Replit Agent that is now in use. Replit also publishes an llms.txt for its documentation and a skills directory where partners publish skills for Agent, so other agents and other companies can build on it.

What we saw and did not count. Replit states a SOC 2 Type 2 attestation; without a public report or scope that covers Agent, it is shown and not counted. Autonomy 4 and Control 2 hold only with options on (connectors, automatic apply, an Enterprise workspace); the card names the level as shipped beside them.

For readers. The customer pages tell a consistent story of people outside engineering building internal tools: Rokt's 700 staff building 135 apps in a day, Leatherman's 119 apps from 147 employees. Those are Replit's figures, on Replit's pages, and are reported here as such.

What would move it

The one piece of public evidence that would lift each card next:

The agent at work

Examples from public pages that name a customer or give figures. Each source was captured and hashed on the date shown.

Plaid

An operations manager built an SLA dashboard with Replit Agent in a hackathon; it is now in use as an internal dashboard. The time savings are the story's own estimates. "How Plaid Built a Production SLA Dashboard with Replit Agent"

Source (captured 2026-11-10T04:21Z, sha256 e86a15bab48e57da) · Replit Agent (Agent 4)

Rokt

700+ employees built 135 internal applications in a 24-hour hackathon. "700+ employees globally built 135 fully functional applications"

Source (captured 2026-11-10T04:21Z, sha256 969b406cd7a30b47)

Leatherman Tool Group

147 employees published 119+ apps since November 2024. "Leatherman: 119 Apps From 147 Employees"

Source (captured 2026-11-10T04:21Z, sha256 dcecb9e86d11fbc2)

Payouts.com

A small team rebuilt its payments platform on Replit; processes tens of thousands of live transactions a month. "rebuilt on Replit with no traditional dev team"

Source (captured 2026-11-10T04:21Z, sha256 ea43865c4b15dbad)

Musixmatch

Adopted across about 120 people; merch tool from prototype to production in about two months. "Musixmatch adopted Replit across about 120 people"

Source (captured 2026-11-10T04:21Z, sha256 1ca29048d44d1f16)

Zillow

Forbes, reposted on Replit's newsroom, reports about 600 Replit seats at Zillow and more than 7,000 apps over a year. "Zillow has roughly 600 Replit seats for its employees"

Source (captured 2026-11-10T05:44Z, sha256 f7aa1e18e16de47b)

The company

The people

Public and professional only, from the company's own pages or a public professional source.

The graph

The company, its people and the stories on the left; each solution, the captured sources behind its card, the indicators those sources evidence and the provisions of the instruments those indicators touch, to the right. "Touched" never means "complies".

people and storiescompanysolutions18 sources28 indicators76 provisions, by instrumentI01: The help page states the model behind Agent runs on Replit's infrastructure and cannot be replaced by the user.I04: The General Agent page states Agent executes code as part of its work and sets up its own environment.I07: In Plan mode Agent brainstorms and plans without changing code or data; the person then chooses to build.I08: The overview states Agent takes actions such as setting up projects and fixing problems.I09: The General Agent FAQ states each chat is contained within a single project.I10: With connectors added, Agent reads and writes to several external services such as BigQuery, Linear, Slack and Notion.I11: The MCP list states Agent connects to external tools and data sources through MCP servers the user installs.I17: With App Testing on, Agent tests the app on its own and hands over to the person at a roadblock such as a login.I18: In the default build mode Agent plans, writes code, sets up infrastructure, tests and deploys on its own from a description.I19: Background tasks wait for review by default; the person can switch on automatic apply of their changes.I20: The overview states what Agent is for: turning plain-language ideas into apps, designs, slides and other outputs.I21: The page defines that Agent reaches external tools only through connectors the user adds and has no access by default.I22: The help page documents known failure cases such as Agent freezing, looping and adding unrequested features, with what to do.I25: On Enterprise an organisation can bring its own OAuth app and set the exact scopes builders can request for a connector.I38: Changes made by background tasks are held for the person to review before they are applied to the main version.I39: The help page states the person can select Stop in the Agent panel to halt Agent immediately.I41: Each checkpoint Agent makes creates a Git commit, a record of what it changed. Documented, so it does not count where an operation fact must be observed.I46: Agent settings such as auto-merge for background tasks and auto-approve for plans are the user's own settings.I47: Workspace custom instructions are guidance in prose that Agent is guided to follow, without a guarantee.I51: The security page states an internal harness continuously probes and tests the AI Agent for safety and security boundaries.I52: The security page names external testing of the platform: a public HackerOne bug bounty and Trail of Bits penetration testing. The source says this covers the platform; it does not name Agent.I57: The Agent help page points users with unresolved Agent problems to Replit Support.I58: The page describes the content used for model improvement: prompts, instructions, context and generated responses.I61: The subprocessor list names the providers of AI models (Anthropic, OpenAI, Google); it lists providers, not which model Agent uses.I68: The documentation index lists a weekly changelog; the latest entry read is dated 6 November 2026.S1: A Forbes article reposted on Replit's newsroom names Zillow, with about 600 seats and more than 7,000 apps its employees made with Replit over a year; the article describes Replit's product as an AI agent that writes its own code. A second source, Replit's Plaid story, names Replit Agent and an internal dashboard in use.S2: The skills directory lists skills published by partners such as Stripe, PayPal and Atlassian for Agent.S3: The docs llms.txt was fetched and resolves, listing Agent, MCP, security and admin pages.ReplitAmjad MasadHaya OdehMichele CatastaLuis Héctor ChávezStacey La TorreMichael GreenwaldScott KennedyJordan WalkeRobert KohseFaris MasadPlaidRoktLeatherman Tool GroupPayouts.comMusixmatchZillowReplit Agent (Agent 4)Replit AgentA4 · R2 · C2 · O2General AgentGeneral AgentPlan modePlan modeTask system / background tasks and worker chatsBackground tasksApp TestingApp TestingRoutinesRoutinesConnectors, custom connectors and MCP servers in AgentConnectors and MCP in AgentReplit MCP Server (Replit in ChatGPT, Claude, Codex, any MCP client)Replit MCP ServerReplit on SlackReplit on SlackAgent Skills, Skills directory and Agent CustomizationAgent SkillsSecurity Agent (Agent security scans) and Auto-ProtectSecurity AgentMonitoring AgentMonitoring AgentPerformance Agent, SEO AgentPerformance and SEO Agentsdocs.replit.com/help/agent-and-aidocs.replit.com/features/agent/ge…docs.replit.com/features/agent/pl…docs.replit.com/features/agent/ov…docs.replit.com/features/mcp/over…docs.replit.com/features/agent/ap…docs.replit.com/features/integrat…docs.replit.com/features/agent/ta…docs.replit.com/features/version-…docs.replit.com/features/agent/ag…docs.replit.com/features/agent/ag…replit.com/securitydocs.replit.com/legal-and-securit…replit.com/subprocessorsdocs.replit.com/_llms/en/features…replit.com/news/funding-announcem…docs.replit.com/features/agent/sk…docs.replit.com/llms.txtIt is a vendor's assistant, used as suppliedI01It generates and executes code with local or cloud privilegesI04It retrieves information and recommends actions; a person actsI07It executes actions itselfI08It works in one domain's dataI09It works across several domains or systemsI10It connects to external tools or services across a trust boundaryI11The agent handles routine tasks; the person is consulted on exceptionsI17The agent works independently; the person approves critical decisions onlyI18The agent has full autonomy within boundaries; the person monitorsI19Its purpose and permitted tasks are stated in writingI20Its autonomy limits and the tools it may use are definedI21Its limits and known failure cases are documented for those who use itI22It asks for narrower access than the whole accountI25High-impact actions need a person's approvalI38There is a way to pause or stop itI39Its actions are loggedI41A limit on it is a setting that its own account could changeI46A limit on it is a rule in proseI47It is tested for safety and security, including by attackI51People who did not build it take part in assessing itI52There is a channel to report problems with it or contest its outputI57The data it uses is describedI58The models and third-party components it uses are listedI61It is improved on a regular cycleI68It is in production: a public source names a customer or gives usage figures for what it does on its ownS1Others build on it: third parties publish integrations, skills or agents for itS2It is machine-readable for agents: it publishes an llms.txt, an AGENTS.md, an API description or an agent cardS3Salesforce 17EAAMM 4OWASP 11MCP authorization 4AAGMM 7Microsoft 9NIST AI RMF 20EU AI Act 4OWASP, AT1 Vendor Embedded AssistantOWASP, AT4 Code Executing AgentSalesforce, Level 1: Information Retrieval AgentsSalesforce, Level 0, how companies can advance from Level 0 to Level 1Salesforce, Level 2: Simple Orchestration, Single DomainSalesforce, Level 1, how to advance from Level 1 to Level 2Microsoft, 200: Repeatable, Business strategySalesforce, Level 0, how companies can advance from Level 0 to Level 1Salesforce, Level 3: Complex Orchestration, Multiple DomainSalesforce, Level 2, how to advance from Level 2 to Level 3Salesforce, Level 3, how to advance from Level 3 to Level 4Microsoft, 400: Capable, Business strategyMicrosoft, 200: Repeatable, Technology and dataOWASP, AT6 Externally Extended AgentNIST AI RMF, GOVERN 6.1NIST AI RMF, MAP 4.1EAAMM, L3 Consultative (survey item Q9)EAAMM, L4 Supervisory (survey item Q9)EAAMM, L5 Autonomous (survey item Q9)Microsoft, 500: Efficient, Business strategyNIST AI RMF, MAP 1.1NIST AI RMF, MAP 2.1NIST AI RMF, MAP 3.3NIST AI RMF, MAP 1.3EU AI Act, Article 13OWASP, Maturity Level 1 Experimentation without Guardrails, key enterprise actionNIST AI RMF, MAP 2.2Microsoft, 300: Defined, Business strategyNIST AI RMF, MEASURE 2.5MCP authorization, Authorization (draft)Salesforce, Level 2, how to advance from Level 2 to Level 3Salesforce, Level 3, how to advance from Level 3 to Level 4AAGMM, GD-02MCP authorization, Authorization (draft)OWASP, Maturity Level 2 Policy-Defined, Human-in-the-Loop, key enterprise actionEU AI Act, Article 14AAGMM, GD-06NIST AI RMF, MAP 3.5Salesforce, Level 3, how to advance from Level 3 to Level 4OWASP, Maturity Level 3 Integrated, Continuous Oversight, key enterprise actionNIST AI RMF, MANAGE 2.4OWASP, Maturity Level 0 Unaware and Ad Hoc, key enterprise actionOWASP, Maturity Level 1 Experimentation without Guardrails, key enterprise actionEU AI Act, Article 12AAGMM, GD-03AAGMM, GD-08OWASP, Maturity Level 0 Unaware and Ad Hoc, key enterprise actionNIST AI RMF, MEASURE 2.6NIST AI RMF, MEASURE 2.7EU AI Act, Article 15NIST AI RMF, GOVERN 4.3NIST AI RMF, MEASURE 1.3NIST AI RMF, MEASURE 3.3NIST AI RMF, GOVERN 5.1Salesforce, Level 1, how to advance from Level 1 to Level 2Salesforce, Level 4, how to maximize valueNIST AI RMF, GOVERN 5.2AAGMM, GD-05EAAMM, Readiness dimension 2Salesforce, Level 1, how to advance from Level 1 to Level 2OWASP, Maturity Level 2 Policy-Defined, Human-in-the-Loop, key enterprise actionAAGMM, GD-12Microsoft, 400: Capable, AI governance and securityMicrosoft, 500: Efficient, Business strategyMicrosoft, 500: Efficient, AI strategy and experienceNIST AI RMF, MANAGE 4.2OWASP, Maturity Level 1 Experimentation without Guardrails, key enterprise actionNIST AI RMF, MEASURE 4.3NIST AI RMF, MAP 1.4Microsoft, 400: Capable, Business strategySalesforce, Level 0, how companies can advance from Level 0 to Level 1Salesforce, Level 4: Multi-Agent OrchestrationSalesforce, Level 4, how to maximize valueAAGMM, GD-09MCP authorization, Authorization (draft)MCP authorization, Authorization (draft)

The evidence

SourceReadSHA-256 (first 16)
docs.replit.com/_llms/en/features.txt2026-11-10T05:44Z2e20971e9962f4ca
docs.replit.com/features/agent/agent-customization2026-11-10T05:44Zb5b576cc757a6d03
docs.replit.com/features/agent/agent-modes2026-11-10T05:44Zd5384b250dce1edd
docs.replit.com/features/agent/app-testing2026-11-10T05:44Z666730392081686e
docs.replit.com/features/agent/general-agent2026-11-10T05:44Z7dee387f587385fc
docs.replit.com/features/agent/overview2026-11-10T05:44Z351087bd1f78f841
docs.replit.com/features/agent/plan-mode2026-11-10T05:44Z1c2834bd959ed8d7
docs.replit.com/features/agent/skills-directory2026-11-10T05:44Za8478b581f6046ad
docs.replit.com/features/agent/task-lifecycle2026-11-10T05:44Zd4df1864788c5069
docs.replit.com/features/integrations/overview2026-11-10T05:44Z29d5d4195ad021dd
docs.replit.com/features/mcp/overview2026-11-10T05:44Z8f0d0d44edb190fb
docs.replit.com/features/version-control/checkpoints-and-rollbacks2026-11-10T05:44Z3a1b8fde4d8a91f3
docs.replit.com/help/agent-and-ai2026-11-10T05:44Zc787756fd19c0948
docs.replit.com/legal-and-security-info/model-improvement2026-11-10T05:44Zb330256ef72ada55
docs.replit.com/llms.txt2026-11-10T05:44Zd2fa30b48300baaa
docs.replit.com/teams/information-security/overview2026-11-10T04:21Z11f1ecd0a4fc3816
en.wikipedia.org/wiki/Replit2026-11-10T04:21Zbb945a741529820a
replit.com/about2026-11-10T04:21Z583a3b3f37782d94
replit.com/customers2026-11-10T04:21Z1b201e019e7fa43d
replit.com/customers/leatherman2026-11-10T04:21Zdcecb9e86d11fbc2
replit.com/customers/musixmatch2026-11-10T04:21Z1ca29048d44d1f16
replit.com/customers/payouts2026-11-10T04:21Zea43865c4b15dbad
replit.com/customers/plaid2026-11-10T04:21Ze86a15bab48e57da
replit.com/customers/rokt2026-11-10T04:21Z969b406cd7a30b47
replit.com/news2026-11-10T04:21Zc05cbc6d2030fe20
replit.com/news/funding-announcement2026-11-10T05:44Zf7aa1e18e16de47b
replit.com/security2026-11-10T05:44Zc69e270c51d716d9
replit.com/subprocessors2026-11-10T05:44Z43bd18511a5644e7

How this report was made

A researcher gathered and captured the sources. Two assessors, working blind to each other, mapped the evidence to indicators; neither assigned a level. A challenger argued against every level three and four and every disagreement. The deterministic scorer computed the levels and the gate checked the rules. A card appears here only after two rounds of review and a sign-off by agents that did not write it (the team's workflow); Replit Agent (Agent 4): reviewed in 2 rounds, signed off by gamma.approve@web-summit.dinis on 2026-11-10. A person's sign-off of the series is still to come.

RoleWho
researchera sub-agent of gamma.studio@web-summit.dinis
assessor Aa sub-agent, source-first
assessor Ba sub-agent on a different model, indicator-first
challengera sub-agent
scorerbuild/spectrum.mjs
analyst, designer, writergamma.studio@web-summit.dinis
reviewgamma.review@web-summit.dinis, 2 rounds
sign-offgamma.approve@web-summit.dinis, 2026-11-10

The vault

Every capture, the dossier, both assessors' mappings, the challenger's notes and this report's data are in a public sgit vault (micbbanr). Read key:

sgit_public_read_a835cecf0324b1147350caf9a8566f08f8c1c10df7eb007bd866eb156f228154:micbbanr

With sgit: sgit clone followed by the key. The key reads; it cannot write.

Corrections

None yet. A correction is made in place, dated here, and carried to the card, the vault and the article.

The article

Replit Agent on the GAMMA-Ray Spectrum: A4 · R2 · C2 · O2, written from this report for PrompTugal.news.