CerebreX Project Example

  ██████╗███████╗██████╗ ███████╗██████╗ ██████╗ ██╗  ██╗
 ██╔════╝██╔════╝██╔══██╗██╔════╝██╔══██╗██╔══██╗╚██╗██╔╝
 ██║     █████╗  ██████╔╝█████╗  ██████╔╝██████╔╝ ╚███╔╝
 ██║     ██╔══╝  ██╔══██╗██╔══╝  ██╔══██╗██╔══██╗ ██╔██╗
 ╚██████╗███████╗██║  ██║███████╗██████╔╝██║  ██║██╔╝ ██╗
  ╚═════╝╚══════╝╚═╝  ╚═╝╚══════╝╚═════╝ ╚═╝  ╚═╝╚═╝  ╚═╝

cerebrex project example — v0.9.4 — every module, live in your browser

forge trace memex auth registry hive kairos ultraplan

01 FORGE live

generate a production MCP server from any OpenAPI spec. one command, ready for Cloudflare Workers.

OPENAPI SPEC URL OR PATH

OUTPUT DIR

parses OpenAPI 3.x and Swagger 2.x specs from URL or local file

generates Zod validation for every tool parameter automatically

wrangler.toml pre-configured for Cloudflare Workers deployment

supports stdio, SSE, and Streamable HTTP transports

ready for cerebrex validate and cerebrex publish

02 TRACE live

real-time MCP call inspector. load a session JSON and walk through every tool call, LLM round-trip, and error step.

drop a trace file here

or click to browse for a JSON trace file
exported with cerebrex trace view --session <id>

browse files

load trace file or drag a .json trace onto the panel above

03 MEMEX live

three-layer persistent memory for agents. KV index (≤25KB, sub-ms reads) → R2 topic blobs (≤512KB) → D1 append-only transcripts (≤1MB). nightly autoDream consolidation via Claude compresses raw history into structured topics automatically.

install the memex mcp server

cerebrex install @arealcoolco/memex-mcp

04 AUTH live

token-based identity and access management for the cerebrex platform. create scoped tokens, manage credentials, authenticate CLI and agent calls.

SIGNED IN AS

—

create token

publish install admin

cli commands

authenticate cli terminal

set token manually terminal

list my tokens terminal

revoke a token terminal

05 REGISTRY live

discover, install, and publish MCP servers. the cerebrex package registry is the central hub for agent tooling.

9 packages in the registry

* official

@arealcoolco/kvstore-mcp

CerebreX KVSTORE — ephemeral key-value session store MCP server for AI agents

mcpcerebrexkvstorev1.0.0

* official

@arealcoolco/datetime-mcp

CerebreX DATETIME — time, date, and timezone MCP server for AI agents

mcpcerebrexdatetimev1.0.0

* official

@arealcoolco/fetch-mcp

CerebreX FETCH — HTTP request MCP server for AI agents

mcpcerebrexfetchv1.0.0

* official

@arealcoolco/hive-mcp

CerebreX HIVE — multi-agent orchestration MCP server

mcpcerebrexhivev1.0.0

* official

@arealcoolco/memex-mcp

CerebreX MEMEX — persistent cloud memory MCP server for AI agents

v1.0.0

* official

@arealcoolco/github-mcp

MCP server for GitHub — repositories, issues, pull requests, commits, users, and search

v1.0.0

* official

@arealcoolco/openweathermap-mcp

MCP server for OpenWeatherMap — current weather, 5-day forecast, and air quality for any location on Earth

v1.0.0

* official

@arealcoolco/nasa-mcp

MCP server for NASA APIs — APOD, Mars Rover Photos, Near-Earth Objects, Earth Imagery, and Image Library

v1.0.0

perf-test-pkg

Performance test package

v1.0.0

06 HIVE live

multi-agent swarm orchestration with three execution strategies: parallel (all agents simultaneously), pipeline (output feeds next), competitive (fastest wins). six built-in presets. every task passes through the risk gate — LOW/MEDIUM/HIGH classification enforced before execution. nothing runs without your consent.

↑ go to auth

07 KAIROS live

autonomous background daemon built on Cloudflare Durable Objects. runs a 5-minute tick loop — continuously checking tasks, monitoring conditions, and taking action without being asked. exponential backoff on errors (1min → 30min cap). every decision written to an append-only D1 audit log.

cerebrex kairos daemon register --agent-id my-agent --goal "Monitor repo activity"

cerebrex kairos task submit --agent-id my-agent --type "data_sync" --payload '{}'

cerebrex kairos log --agent-id my-agent --limit 20

5-minute tick loop via Durable Object alarm — no polling, no cron, always on

exponential backoff on errors — 1min → 2 → 4 → up to 30min cap, resets on success

append-only D1 audit log — every decision, reasoning, and action persisted

JSON-validated tick responses — malformed output is rejected before acting

50KB goal size limit — prevents runaway context injection

strict agentId validation — alphanumeric format enforced, injection blocked

08 ULTRAPLAN live

Opus-powered long-range planning with human-in-the-loop approval. Claude Opus decomposes your goal into a structured execution plan → you review and approve → HIVE executes all subtasks in parallel. goals capped at 50KB. every subtask inherits HIVE's risk gate. full audit trail in D1.

generate a plan → approve → execute

cerebrex ultraplan "Build a REST API with auth, docs, and tests"

cerebrex ultraplan execute --plan-id <id> --approve

Claude Opus decomposes complex goals into structured, executable subtask trees

human approval gate — the plan requires explicit confirmation before any execution begins

parallel HIVE execution — approved tasks fan out across the swarm simultaneously

50KB goal limit — enforced at API level, prevents context injection

risk gate inheritance — every subtask passes through HIVE's LOW/MEDIUM/HIGH classification

full auditability — plan, approval decision, and execution results all logged to D1