Rippletide adds an authority layer that validates, constrains, or blocks agent actions at runtime before they impact real systems or customers.

Web Platform · GitHub · Documentation · Discord

This repository is your starting point for Rippletide — Eval, Context Graph MCP, and Coding Agents.

---

Table of Contents

• What is Rippletide?
• Rippletide Platform

Core Modules:

#	Module	What it does
1	Agent Evaluation - CLI	Validate before you ship
2	Context Graph - MCP	Give your agents persistent memory across sessions
3	Coding Agents	A persistent memory layer for Claude

_{Decision Runtime — Enterprise Only. Build deterministic agents with less than 1% hallucination rate. Contact us to learn more.}

---

What is Rippletide?

Rippletide adds an authority layer that validates, constrains, or blocks agent actions at runtime before they impact real systems or customers.

	Without Rippletide	With Rippletide
Hallucinations	Variable, hard to control	Less than 1% by design
Memory	Lost between conversations	Persistent context graph
Guardrails	Prompt-based, easy to bypass	Engine-level, 100% compliance
Explainability	Black box	Every decision is traceable
Evaluation	Manual spot checks	Automated, CI-ready testing

---

Agent Evaluation CLI

Eval is the entry point to Rippletide. Before adding memory or decision runtime, start by testing what your agent already does. Plug your agent into our CLI and Rippletide auto-generates test questions to evaluate its responses. Rippletide spots hallucinations by fact-checking each output and suggests improvements when sources are missing.

When to use it: Before every deployment, in CI pipelines, and during development to catch regressions.

Installation

Install globally via npm:

npm install -g rippletide

Or use directly with npx:

npx rippletide

Quick Start

Simply run:

rippletide

You'll be prompted for:

1. Agent endpoint — Your API URL (e.g. http://localhost:8000)
2. Knowledge source — Choose between files, Pinecone, or PostgreSQL

The CLI will then:

• Load your test questions
• Send them to your agent
• Show real-time progress
• Display evaluation results with pass/fail and justifications

Command Line Options

rippletide eval [options]

Option	Description	Example
-t, --template <name>	Use a pre-configured template	rippletide eval -t banking_analyst
-a, --agent <url>	Agent endpoint URL	rippletide eval -a localhost:8000
-k, --knowledge <source>	Knowledge source: files, pinecone, or postgresql	rippletide eval -k pinecone
--debug	Show detailed error information	rippletide eval --debug
-h, --help	Show help message	rippletide --help

Data Source Options

Local Files (default):

rippletide eval -a localhost:8000

Reads Q&A pairs from qanda.json in the current directory.

Pinecone:

rippletide eval -a localhost:8000 -k pinecone \
  -pu https://db.pinecone.io \
  -pk pcsk_xxxxx

PostgreSQL:

rippletide eval -a localhost:8000 -k postgresql \
  -pg "postgresql://user:pass@localhost:5432/db"

Custom Endpoint Options

For non-standard APIs:

rippletide eval -a localhost:8000 \
  -H "Authorization: Bearer token, X-API-Key: key" \
  -B '{"prompt": "{question}"}' \
  -rf "data.response"

Option	Description
-H, --headers	Custom headers (comma-separated)
-B, --body	Request body template (use {question} placeholder)
-rf, --response-field	Path to response in JSON (dot notation)

Templates

Pre-built configurations for common agent use cases:

Template	Description
banking_analyst	Financial Q&A agent
customer_service	Support agent testing
blog_to_linkedin	Content repurposing agent
luxe_concierge	Luxury services agent
local_dev	Local development agent
openai_compatible	OpenAI-compatible endpoints
project_manager	Project management agent

rippletide eval -t customer_service

→ Full Evaluation docs

---

Context Graph - MCP

A persistent memory layer for your AI agents. Connect any MCP-compatible client (Cursor, Claude Desktop, Claude Code) and your agent can remember facts, decisions, and context across sessions.

Quick Start

Add this to your MCP client config:

{
  "mcpServers": {
    "rippletide": {
      "type": "url",
      "url": "https://mcp.rippletide.com/mcp?agentId=your-agent-id"
    }
  }
}

Get your agentId from the Rippletide platform.

Client	Config location
Cursor	~/.cursor/mcp.json
Claude Desktop	MCP settings in the app
Claude Code	.mcp.json at project root

→ MCP docs

---

Coding Agents

Give Claude Code a shared, persistent memory. Store your team's conventions once — naming rules, architecture decisions, error handling policies — and every Claude session pulls from the same source automatically.

Quick Start

npx rippletide-code

One command to authenticate, scan your repo, select rules, and install hooks. Every Claude Code session in this project will have access to your rules from that point on.

If you need to point the CLI at a specific coding-agent backend, set one base URL and the CLI will derive the related endpoints from it:

RIPPLETIDE_API_URL="https://coding-agent.up.railway.app" npx rippletide-code

Setup Flow

The CLI now has two setup flows.

Individual

• Start the CLI and choose 1. Individual workspace
• The CLI uses the standard Rippletide cloud login flow
• After login, rules sync with the individual workspace

Enterprise

• Point the CLI at a client-hosted coding-agent backend:

RIPPLETIDE_API_URL="https://company-coding-agent.internal" npx rippletide-code

• Start the CLI and choose 2. Enterprise backend
• The CLI stores that backend as the active coding-agent backend for later runs
• The enterprise backend handles uploads, rule extraction, and Anthropic-backed processing

In practice:

• Individual = Rippletide cloud auth + cloud workspace
• Enterprise = company backend + enterprise-local coding-agent flow

For the maintainer release flow for the internal CLI channel, see context-graph/CLI_RELEASE_FLOW.md.

Features

Feature	What it does
Rule enforcement	Rules are injected into every prompt. Code that violates a rule is blocked before it hits the file. Claude auto-rewrites until it complies.
Rule management	Add, edit, or delete rules in natural language. No config files. Changes take effect immediately.
Rule sharing	Send your rule set to a colleague with invite-rules <email>. They type receive-rules <otp> and get a conflict report with new rules, duplicates, and contradictions.
Planning	/plan generates an implementation plan and reviews it against your rules. Violations are revised automatically until the plan passes.
Team governance	Create a team, push your rules, and have every engineer sync from the same source. Read-only mode lets engineers inherit standards without modifying them.

Commands

Action	Command
Share rules	invite-rules <email>
Receive rules	receive-rules <otp>
Create team	create-team <name>
Join team	join-team <name> [approver_email]
Approve member	approve-join <team> <otp> <email>
Push to team	push-rules <team>
Sync from team	sync-rules <team>
Read-only connect	npx rippletide-code --read-only

Data privacy

Rippletide only relies on the context available inside your local Claude Code workflow — your current chat session and your CLAUDE.md. Your codebase is not centrally analyzed. Rippletide stores only the extracted rules, not project content.

→ Coding Agents docs

---

Rippletide Platform

The Rippletide Platform brings everything together. Build agents without writing code, connect your knowledge sources, set guardrails that the LLM cannot override, and see exactly how your agent reasons through every decision — all in one place.

• Visual Agent Builder — configure agents without code
• Knowledge Connectors — import from Amazon Bedrock, PDFs, or manual Q&A
• Knowledge Visualization — see your agent's full knowledge graph
• Guardrail Configuration — rules enforced at runtime, not in the prompt

---

Repository Structure

rippletide/
├── agent-evaluation/       # TypeScript CLI for agent evaluation
│   ├── bin/rippletide      # CLI entry point
│   ├── src/                # Source (api, components, errors, utils)
│   └── templates/          # Pre-built agent configs
├── context-graph/          # Rust MCP server for coding agents
│   ├── src/                # Rust source
│   └── npm/                # Multi-platform binary packages
└── docs/                   # Documentation site (Mintlify)

---

Development

git clone https://github.com/rippletideco/rippletide.git

# Agent Evaluation CLI
cd rippletide/agent-evaluation
npm install
npm run build
npm run eval         # run development version

# Context Graph MCP server
cd rippletide/context-graph
cargo build --release

---

Contributing

We welcome contributions. Please read our Contributing Guidelines, Code of Conduct, and Security Policy before opening a PR.

---

Support

• Discord: Join our community
• GitHub Issues: Report bugs
• Docs: docs.rippletide.com

---

Built with ❤️ by the Rippletide team