XRevery
Log in Sign up free
Team-scale agentic development

Run a team of AI agents on your repos. In parallel. Under your rules.

XRevery turns a brief into reviewed pull requests: spec, test, implement and verify in isolated cloud sandboxes. Use platform credits or your own model subscriptions and API keys. Human gates wherever you want them.

Sign up, no credit card See how it works

Up to $100 in token spend included on the free tier.

Runs in flight 6 sandboxes
api-gateway PR #418
billing-svc Needs you
web-app Testing
infra-tf Review
sdk-python PR #77
docs-site Spec
AnalyseSpecImplementVerify $4.18 this run group
Works with GitHubClaudeAWS BedrockGoogle GeminiCodexHugging FaceOpenAI-compatible endpointsMCP
01 / The harness already exists

The AI engineering platform you would otherwise spend many months building.

Most teams are still assembling their own harness from scripts, prompts and hope. XRevery ships it as product: multi-agent workflows, a Coordinator that plans and routes, token and context discipline, and a catalogue you can build on.

Cohort

Specialists, not one generalist

Spec writer, reviewer, coder and tester work in sequence, with optional security and holistic reviewers on the work that warrants them.

Coordinator

Plan in conversation

Talk through the task, attach context, then confirm to create the run. Routine work gets handled directly; the rest is escalated to a full pipeline.

Catalogue

Agents, workflows, skills, MCP

Start from verified building blocks instead of a blank prompt file, then adapt them to how your organisation actually works.

Coordinator
Rate limiting on the public API is inconsistent between the gateway and billing service. Make it uniform, keep the current header names, and add tests.
Two repos affected. I will run the Standard workflow on both, share one spec, and hold a human gate before implementation. Estimated spend $3.10 to $5.40.
Confirm and create run Edit plan
Workflow shapes
Direct One-shot changes and small fixes
Light Spec, implement, verify
Standard Full cohort with review and tests
Composed Coordinator builds the graph for the task
Workflows & agents →
02 / Your laptop is not the bottleneck

Ten agent runs in flight. Zero tabs open on your machine.

Every run gets its own isolated cloud sandbox on pooled runners. Start work, close the laptop, and get pulled back in only when a decision is actually yours to make.

Parallel by default. Fan a brief across several repos, or run unrelated tasks side by side without fighting over one machine.

Full timeline per run. Live activity, a run profiler and a cost ledger with a token breakdown, so nothing about a run is a black box.

Ends in a pull request. One PR per repo, on a branch, and runs can wait on CI before they count as done.

Cost ledger $4.18
Implement1.9M tokens
Review1.1M tokens
Tests0.7M tokens
Billed toYour Anthropic key
Needs your decision

billing-svc wants to change the public error shape for 429 responses. Approve, or keep the current contract?

Approve Keep contract Comment
03 / It learns your business, not just your repo

Every run makes the next one smarter about your stack.

Repo profiles capture how each codebase is built. A System Wiki grows from your code and from finished runs. Systems group related repos so a single brief can move across service boundaries without re-explaining your architecture every time.

Repo profile
stackGo 1.22, Postgres
testsgo test ./...
layouthexagonal
owners@platform
ci gaterequired
+
System wiki
How rate limiting works across services
Auth token lifecycle and refresh rules
Why billing owns invoice numbering
Updated by run #418, 2 days ago
Coding turn

The coder starts with your conventions already in context, not after four exploratory reads of the repo.

context loaded: profile + 3 wiki pages
redundant discovery avoided: 41k tokens
04 / Engineers stay in charge

Agents do the grind. Your team owns architecture, design and the merge button.

XRevery is built to elevate engineers, not to route around them. Put a human checkpoint anywhere it matters: after the spec, before implementation, or on anything touching a sensitive path. Agents open pull requests; people merge them.

Written for engineering teams
Where humans sit in the pipeline
Analyse and spec
Agent drafts the change and the test plan
You approve the approach HUMAN
Optional, and you choose which tasks need it
Implement, test, review
Security and holistic reviewers can be added per workflow
Pull request opened, CI runs
Address PR comments in a follow-up run
Your team merges HUMAN
Always. Agents never own the merge.
05 / Control of models, cost and data

Your models, your tokens, your rules.

Connect the provider keys and model subscriptions you already pay for. Spend routes to your allowances where you configure it, with clear attribution in Usage. Or stay on platform credits and never touch a key.

Bring your own tokens

Anthropic, OpenAI, AWS Bedrock, Google Gemini, Hugging Face, and custom OpenAI-compatible endpoints.

Claude Bedrock Gemini Codex Hugging Face Open source routes + your endpoint

Caps before surprises

Routing policy decides which model does what. Spend caps stop a run group before it runs away.

Monthly cap$2,000
Used$761 · 38%
On your keys$540

Where it runs is your call

Managed SaaS, or sandboxed on-prem and VPC deployment when your code cannot leave your estate.

Isolated sandbox per run
Per-organisation isolation
Managed secrets, scoped to a run
Roles, invites and OIDC login
Control & trust →

Not a copilot in your editor.

If you want an assistant while you type, keep the one you have. XRevery is for work you want to hand over as a brief and get back as a reviewed pull request.

A do-it-yourself harness Productised workflows, gates, catalogue and operations, maintained for you.
A single IDE agent Team runs across multiple repos, the whole pull request lifecycle, and shared budgets.
A CLI on a laptop Cloud sandboxes, real parallelism, and usage you can see for the whole organisation.
A generic agent framework An opinionated software delivery cohort, aimed at production codebases from day one.