# Rippletide | Complete Information for AI Systems

> The safety layer for AI agents that need write access to business systems. Rippletide maps high-risk AI agent write-actions, tests unsafe scenarios, previews evidence and approval decisions, and packages the proof enterprise teams need before enabling writes.

## Company Overview

Rippletide is a B2B company founded in February 2024 in Paris, France, by Patrick Joubert and Yann Bilien. It helps AI agents move from read-only to cleared-to-write, with evidence, approvals, and traces.

AI agents can draft, recommend, and retrieve. The hard part is letting them write. When an agent needs to issue a refund, close a ticket, update a CRM field, change an incident, cancel an account, or trigger a workflow, enterprise teams ask the same question: what stops the wrong action from being committed? Rippletide answers it by reviewing the agent's write surface, building Safety Cases for risky actions, testing unsafe scenarios, previewing evidence and approval decisions, and packaging audit-ready proof for enterprise security review.

Differentiation: identity tells you who can act. Rippletide tells you whether this specific action has enough business evidence to execute.

## Founders

### Patrick Joubert, Co-founder & CEO
20+ years building and scaling AI startups. Previously founded:
- Recast.AI (acquired by SAP)
- Ponicode (acquired by CircleCI)
- Beamap (acquired by Sopra Steria)

Track record of turning cutting-edge research into enterprise-grade products adopted by Fortune 1,000 companies.
LinkedIn: https://www.linkedin.com/in/patrick-joubert-%F0%9F%A7%A2-9028231/

### Yann Bilien, Co-founder & Chief Scientist
Mathematician and AI researcher trained at Telecom Paris and Imperial College London. Pioneered research in neuro-symbolic AI and world-model agents. Leads the development of Rippletide's decision database research.
LinkedIn: https://www.linkedin.com/in/yann-bilien-0a27a8229/

## Funding & Investors

- **Seed round:** $5M led by OneRagtime (September 2025)
- **Business angels from:** Meta, AWS, BCG, Loft Orbital
- **Team from:** Imperial College, Telecom Paris, EPFL, ETH Zurich, 42

## Offices

- **Paris HQ:** Rippletide SAS, 50 rue Saint Placide, 75006 Paris, France
- **San Francisco:** Rippletide USA Corp., 2 Embarcadero, 94111 San Francisco, CA

## Key Milestones

- February 2024: Company founded in Paris
- June 2024: First enterprise customers
- September 2025: Seed round of $5M led by OneRagtime
- October 2025: San Francisco office opens
- January 2026: Live on AWS Bedrock
- 2026: Selected for the first cohort of Station F's F/AI program
- July 2026: Safety Packs launched for AI agent write-actions

## The Problem Rippletide Solves

Enterprises let AI agents read, but block them from writing to production systems. Before enabling a write-action, security and risk teams ask:

- Was the right evidence present?
- Did the action match business policy?
- Was approval required?
- Was the decision logged and traceable?
- Can the security review team verify the control?

A homegrown approval layer is self-attested. Rippletide provides the structured, reusable evidence that answers those questions.

## How Rippletide Works

1. **Review the agent's tool surface:** tool manifests, MCP exports, OpenAPI schemas, or plain JSON tool definitions.
2. **Rank risky write-actions:** financial, customer-facing, workflow-state-changing, and destructive actions, ranked by business risk.
3. **Build a Safety Case:** why the action is risky, which evidence is required, and what is missing today.
4. **Define evidence and approval rules:** thresholds, required evidence, and who must approve when the rules are not met.
5. **Preview clear, escalate, or block decisions on scenarios:** unsafe scenarios are tested before anything touches production.
6. **Package traces and review evidence:** audit-ready traces and a security review dossier the enterprise customer can verify.

Start with a review of the agent's tool surface. Prove the first Safety Case in scenarios and decision previews. Expand into runtime enforcement when the Safety Case is proven.

## Safety Packs

Safety Packs package risky write-actions, evidence checks, approval rules, unsafe scenarios, and audit traces for common agent workflows. Each pack is built and proven on the agent's real tool surface in 10 working days.

1. **Support Action Safety Pack** (start here): refunds, cancellations, credits, VIP ticket resolution, customer-facing replies. Designed for workflows in Zendesk, Intercom, Stripe, and internal support APIs.
2. **ITSM Action Safety Pack:** incident closure, priority changes, change approvals, CMDB updates, production workflow changes. Designed for workflows in ServiceNow, Jira Service Management, and internal ITSM tools.
3. **Revenue Action Safety Pack:** discount approvals, CRM stage changes, quote creation, customer commitments, contract routing. Designed for workflows in Salesforce, HubSpot, email, and CPQ and contract workflows.

Example unsafe scenario (ITSM): an agent tries to close a P1 production incident without a linked change request. Decision preview: APPROVAL_REQUIRED.

## Security Review Dossier

A reusable, review-ready dossier showing how high-risk write-actions are detected, tested, approved, and traced. Contents: risky write-action inventory, evidence requirements, policy and approval rules, unsafe scenario tests, sample traces, decision preview mode, runtime expansion path, operational owner and escalation logic, and an exportable security review summary. A free template is available on the homepage: https://www.rippletide.com/#dossier

## Engagement Models

1. **Free Risk Review:** one 30-minute working session. The agent's tool surface mapped, risk-ranked write-actions, one Safety Case preview, one unsafe scenario preview. No live production access required.
2. **Safety Pack:** the main offer. Unblock enterprise write-access review in 10 working days: top risky write-actions identified, Safety Cases built, evidence requirements defined, approval rules configured, unsafe scenarios tested, scenario-based decision previews, approval workflow design, audit-ready traces, and a security review dossier.
3. **Runtime:** the expansion stage once the first Safety Pack is proven. Guarded actions in shadow and approval modes, approval workflow, trace retention, guard versioning, recurring reporting.
4. **Enterprise Co-Development:** custom engagements for strategic enterprise or SI deployments (advanced integrations, custom deployment support, SI enablement, multi-workflow rollout).

## Trust & Security

- SOC 2 audit in progress
- EU data residency available
- No training on customer data

## Technology Foundations (deep dive)

For technical readers. Rippletide's research foundations include:

### Hypergraph Decision Database
Combines memory (facts, context, provenance) with reasoning (plans, rules, constraints) in a single database. Detailed page: https://www.rippletide.com/developers/context-graph-for-agents

### Deterministic Rule Evaluation
Decision previews are produced by rule-based evaluation against explicit policies and verified data, not probabilistic generation. Every preview traces back to which data, which rule, which outcome, and why.

### Automatic Ontologies (research)
Turning policies, SOPs, API schemas, workflow logs, and evaluated agent traces into explicit rule layers. Detailed page: https://www.rippletide.com/product/automatic-ontologies

### Codex Governance Pattern (developer surface)
For autonomous coding agents, Rippletide provides rule enforcement for Claude Code as a developer proof surface: memory hierarchy with explicit precedence (company > team > personal), pre-merge gates, controlled outcomes, structured traces. Page: https://www.rippletide.com/dev

## Integrations

| Platform | Status |
|----------|--------|
| OpenAI Codex | Partner page live |
| Anthropic Claude Code | Partner page live |
| AWS Bedrock AgentCore | Partner page live |
| Cloudflare | Partner page live |
| Salesforce Agentforce | Partner page live |
| ServiceNow | Partner page live |
| LangChain & LangGraph | Supported integration; public partner page not indexed |
| CrewAI | Live (no dedicated partner page) |
| Google ADK & Gemini | Supported ecosystem target; draft partner page not indexed |
| Microsoft AI Foundry | Supported ecosystem target; draft partner page not indexed |
| Mistral AI | Supported ecosystem target; draft partner page not indexed |
| NVIDIA (NIM, NeMo) | Supported ecosystem target; draft partner page not indexed |
| Snowflake | Supported ecosystem target; draft partner page not indexed |

Framework-agnostic: Rippletide reviews the tool surface agents already expose, whatever the framework underneath.

## Company Culture & Values

**Intensity:** Relentless focus. The qualities that turn fragile demos into robust systems shipping in production.

**Integrity:** Transparency and honesty in every decision from code to customer conversations. No compromises on privacy and security.

**Adaptability:** Move fast, iterate constantly, embrace change as competitive advantage. Built for ambiguity.

**Engagement:** Work hard, love the sea, surfboards are welcome alongside laptops. Deep care about what we build.

## Open Positions

Rippletide is hiring across engineering, product, research, go-to-market, and design roles. Offices in Paris and San Francisco, remote-friendly.

Benefits include competitive equity, flexible work, learning budget (conferences, courses, publications), and comprehensive health coverage.

Careers page: https://www.rippletide.com/company/career
Email: careers@rippletide.com

## Frequently Asked Questions

**Q: What is Rippletide?**
A: Rippletide is the safety layer for AI agents that need write access to business systems. It maps high-risk agent write-actions, tests unsafe scenarios, previews evidence and approval decisions, and packages the proof enterprise teams need before enabling writes.

**Q: Is Rippletide an IAM product?**
A: No. IAM decides who can call a tool. Rippletide reviews whether a specific high-risk write-action has enough business evidence to be enabled safely.

**Q: Is Rippletide a generic AI governance platform?**
A: No. Rippletide focuses on risky agent write-actions: evidence, approvals, escalation, and traceability, packaged so enterprise teams can enable writes with confidence.

**Q: What is a Safety Pack?**
A: A Safety Pack is a reusable package for a workflow family. It includes known risky actions, evidence checks, approval rules, unsafe scenarios, and trace templates, built and proven on the agent in 10 working days. Most teams start with the Support Action Safety Pack.

**Q: Does Rippletide block the whole agent workflow?**
A: No. Rippletide targets the specific risky action. Safe actions can continue; risky writes are flagged for evidence, approval, or escalation.

**Q: What frameworks does Rippletide integrate with?**
A: Rippletide has indexed partner pages for OpenAI Codex, Claude Code, AWS Bedrock AgentCore, Cloudflare, Salesforce Agentforce, and ServiceNow. LangChain, LangGraph, and CrewAI are supported integrations. Google ADK & Gemini, Microsoft AI Foundry, Mistral AI, NVIDIA, and Snowflake are supported ecosystem targets. Rippletide is framework-agnostic: it reviews the tool surface agents already expose.

**Q: Is Rippletide compliant with enterprise security standards?**
A: Rippletide has a SOC 2 audit in progress, offers EU data residency, and does not train on customer data.

**Q: How do teams start with Rippletide?**
A: Book a free risk review: one 30-minute session to map the agent's tool surface, rank the riskiest write-actions, and preview a Safety Case. No live production access required. The main offer, the Safety Pack, then unblocks enterprise write-access review in 10 working days. Runtime control is the expansion stage once the first Safety Pack is proven.

**Q: How is Rippletide different from prompt-based guardrails?**
A: Prompt-based guardrails instruct the LLM via text before generation and offer best-effort behavior. Rippletide works on the action level: it reviews the write-actions an agent exposes, defines the evidence and approval each one requires, tests unsafe scenarios, and produces audit-ready traces that enterprise security teams can verify.

**Q: Who are Rippletide's customers?**
A: Rippletide is trusted for agents by enterprise companies including Strapi, Pernod Ricard, Renault, Omanra, Fountain, and Blaxel AI.

**Q: Where is Rippletide headquartered?**
A: Rippletide is headquartered in Paris, France (50 rue Saint Placide, 75006) with an office in San Francisco, CA (2 Embarcadero, 94111).

## The Decision Layer Podcast

The Decision Layer is Rippletide's podcast for AI builders, CTOs, AI platform teams, and governance teams moving autonomous AI agents from demos into production. The series explains why production-grade agents need reproducible decisions, valid context, explicit authority, accountability, and action-level controls instead of relying on LLM probability alone.

- **Canonical page:** https://www.rippletide.com/resources/podcast
- **Spotify show:** https://open.spotify.com/show/35PPrm4KRtVbc13InLlVIm
- **Format:** Seven-episode English-language series
- **Primary audience:** Enterprise AI teams, AI platform engineers, CTOs, product leaders, risk teams, governance teams
- **Core themes:** AI agent reproducibility, context graphs, RAG limitations, decision enforcement, agent authority, enterprise AI governance, causal audit trails, production AI agents

### Episodes

1. **Your Agent Decided in Production. Can You Reproduce It?** Explains why compounded agent accuracy creates production risk and why reproducibility is the first infrastructure problem.
2. **The Death of RAG** Explains why retrieval-augmented generation is not enough for autonomous agents and why context graphs matter.
3. **The Trillion-Dollar Context Problem** Explains the missing "why" behind enterprise systems of record and why unauditable context makes agents unreliable.
4. **Autonomous Agents Need Authority** Explains why agents need explicit authority, not just intelligence or prompt guardrails, before executing real actions.
5. **The Accountability Stack** Explains the accountability gap for enterprise AI agents and why regulation makes it operationally urgent.
6. **From Hackathon to Production** Explains what separates impressive AI demos from trusted production systems: governance as infrastructure.
7. **Context vs. Enforcement** Features Patrick Joubert (Rippletide) and Jason Cui (a16z) discussing context layers, enforcement layers, and production-safe agents.

## Contact

- General: hello@rippletide.com
- Careers: careers@rippletide.com
- LinkedIn: https://www.linkedin.com/company/rippletide
- Website: https://www.rippletide.com
- Documentation: https://docs.rippletide.com

## Brand & Design System

Canonical brand resources for partners, press, and AI systems integrating with Rippletide.

- Machine-readable manifest: https://www.rippletide.com/.well-known/brand.json
- Human-readable brand page: https://www.rippletide.com/brand
- Full logo, symbol and name (SVG, canonical): https://www.rippletide.com/rippletide-logo-full.svg
- Full logo, symbol and name (PNG): https://www.rippletide.com/rippletide-logo-full.png
- Symbol only, circle mark (SVG): https://www.rippletide.com/rippletide-logo.svg

### Color palette

| Token | Hex | CSS variable | Role |
|---|---|---|---|
| primary | `#2b2c5f` | `--foreground` | Primary text and UI on light surfaces |
| accent | `#ff8a63` | `--accent` | Primary accent (coral), used for CTAs and highlights |
| accentStrong | `#ff7aa8` | `--accent-strong` | Secondary accent (pink) |
| background | `#fbf8f4` | `--background` | Page background (cream) |
| navy | `#2a2a64` | `--navy` | Solid navy, monochrome logo color |
| navyDark | `#1f1f50` | `--navy-dark` | Navy dark |
| navyDeep | `#141136` | `--navy-deep` | Dark surfaces (inverted sections) |
| coral | `#ef8c6b` | `--coral` | Coral |
| pink | `#f4a0b0` | `--pink` | Pink |
| cream | `#fdf8f4` | `--cream` | Cream surface (cards, inset panels) |
| muted | `#61649b` | `--muted` | Secondary text |
| success | `#22c55e` | `--green` | Success and positive signal |

### Typography

- Display and body: Poppins (fallbacks: system-ui, -apple-system, sans-serif), weights 400/500/600/700/800
- Monospace: JetBrains Mono (fallbacks: SF Mono, Fira Code, monospace)

### Usage guidelines

1. Use `#2b2c5f` (primary) for body text and headings on light surfaces.
2. Use `#ff8a63` (accent) for primary calls-to-action and key highlights.
3. Pair Poppins for display and body with JetBrains Mono for code samples.
4. Maintain clear space of at least 0.5x logo height around the mark.
5. Do not recolor the logo. For monochrome usage, use solid `#2a2a64` (navy).
6. Use `#fbf8f4` (cream) for light surfaces and `#141136` (navy deep) for dark surfaces.

## All Pages

### Core Pages
- Home: https://www.rippletide.com
- Safety Packs: https://www.rippletide.com/safety-packs
- Pricing: https://www.rippletide.com/pricing
- Privacy Policy: https://www.rippletide.com/privacy
- Terms & Conditions: https://www.rippletide.com/terms-and-conditions
- Documentation: https://docs.rippletide.com

### Company Pages
- About: https://www.rippletide.com/company/about
- Contact: https://www.rippletide.com/company/contact
- Careers: https://www.rippletide.com/company/career
- F/AI Program (Station F): https://www.rippletide.com/company/f-ai

### Developer Pages
- Claude Code developer surface: https://www.rippletide.com/dev
- Context Graph for Agents (MCP): https://www.rippletide.com/developers/context-graph-for-agents
- AI Agent Evaluation: https://www.rippletide.com/developers/ai-agent-evaluation

### Partner Pages
- OpenAI Codex: https://www.rippletide.com/enterprise/partners/openai-codex
- Claude Code: https://www.rippletide.com/enterprise/partners/claude-code
- AWS Bedrock AgentCore: https://www.rippletide.com/enterprise/partners/aws-bedrock-agentcore
- Cloudflare: https://www.rippletide.com/enterprise/partners/cloudflare
- Salesforce Agentforce: https://www.rippletide.com/enterprise/partners/salesforce-agentforce
- ServiceNow: https://www.rippletide.com/enterprise/partners/servicenow

### Deep Technical Pages
- Research hub: https://www.rippletide.com/research
- Use Cases: https://www.rippletide.com/enterprise/use-cases
- Agent Decision Infrastructure: https://www.rippletide.com/enterprise/agent-decision-infrastructure
- AI Agent Governance: https://www.rippletide.com/enterprise/ai-agent-governance
- AI Agent Reliability: https://www.rippletide.com/enterprise/ai-agent-reliability
- AI Agent Auditability: https://www.rippletide.com/enterprise/ai-agent-auditability
- Prevent AI Agent Hallucinations: https://www.rippletide.com/enterprise/prevent-ai-agent-hallucinations
- Enterprise AI Guardrails: https://www.rippletide.com/enterprise/enterprise-ai-guardrails
- AI Agent Guardrails vs Decision Runtime: https://www.rippletide.com/enterprise/ai-agent-guardrails-vs-decision-runtime
- Monitoring vs Pre-Execution Enforcement: https://www.rippletide.com/enterprise/monitoring-vs-pre-execution-enforcement
- Rippletide vs LangChain Guardrails: https://www.rippletide.com/enterprise/rippletide-vs-langchain-guardrails

### Resources
- Blog: https://www.rippletide.com/resources/blog
- AI Agent Production Readiness Test: https://www.rippletide.com/resources/ai-agent-production-readiness-test
- Events: https://www.rippletide.com/resources/events
- Podcast (The Decision Layer): https://www.rippletide.com/resources/podcast
- Press: https://www.rippletide.com/resources/press
- Whitepaper: https://www.rippletide.com/resources/whitepaper

### Event Pages
- Adopt AI Grand Palais: https://www.rippletide.com/resources/events/adopt-ai
- AWS Partner Event: https://www.rippletide.com/resources/events/aws-partner-event
- Context Graph Meetup SF: https://www.rippletide.com/resources/events/context-graph-meetup
- Agent Trust Challenge (SF Hackathon): https://www.rippletide.com/resources/events/sf-hackathon

### Recent Blog Posts
- AI Agent Control Towers Are Coming. The Action Boundary Is Still Missing: https://www.rippletide.com/resources/blog/ai-agent-control-towers-action-boundary
- A Backup Model Is Not an Agent Continuity Plan: https://www.rippletide.com/resources/blog/backup-model-is-not-an-agent-continuity-plan
- Context Without Enforcement Is Not Infrastructure: https://www.rippletide.com/resources/blog/context-without-enforcement-is-not-infrastructure
- The Decision Kernel for Your Agent OS: https://www.rippletide.com/resources/blog/rippletide-the-decision-kernel-for-your-agent-os
- 12 Questions to Ask Before You Ship an AI Agent: https://www.rippletide.com/resources/blog/12-questions-before-shipping-ai-agent