Rinkata

Product Introduction

1. Definition: Rinkata is an agent-agnostic AI product alignment hub and shared memory layer for software teams. It connects product intent from idea to deploy through the Rinkata Hub UI and MCP-enabled read/write access for AI coding agents. Core entities include Goals, specs, tickets, Decisions, and completion evidence such as tests, screenshots, and pull requests. It integrates with Claude, Codex, Cursor, ChatGPT, Gemini, and Grok, and provides a rinkata CLI plus agent integrations. It is not a code generator, IDE, or repository replacement; it is the alignment layer between humans, agents, and project truth. Product URL: https://rinkata.dev.
2. Core Value Proposition: Rinkata prevents intent drift in AI-assisted software delivery by keeping what you meant to build, what you committed to build, and what you actually deployed connected in one source of truth. It lets agents search project knowledge before inventing context, routes judgment calls to humans, and automatically attaches proof to completed work. Primary keywords: AI agent alignment, MCP shared memory, product intent management, spec-driven development, agentic coding, human-in-the-loop decisions, completion evidence, knowledge write-back.

Main Features

1. Intent Thread: Idea → Goal → Work → Decision → Evidence: Rinkata Hub UI models delivery as a connected intent thread. Ideas can be captured from a notebook or extracted from documents, then promoted to Goals with acceptance criteria before agents start. Work is claimed from those Goals, with specs and Knowledge attached. When an agent hits a judgment call, the issue surfaces as a Decision under Needs attention, where humans Accept or Reject. Completed tickets produce Evidence packs that trace back to the Goal and the decision history. This creates end-to-end traceability from product intent to deploy.
2. Shared Knowledge Hub with MCP Read/Write Memory: Teams upload designs, PRDs, Figma exports, architecture notes, and product briefs into searchable Knowledge areas and documents. Agents read and write project truth over MCP, so Claude, Codex, Cursor, ChatGPT, Gemini, and Grok can retrieve covering docs before acting. The system stores graph artifacts such as Goals, specs, and tickets. Completed work writes itself back into Knowledge automatically, reducing manual documentation and stale docs.
3. Agent Work Board with Human Judgment Gates: The Work board shows progress for all active work, including agent-claimed Goals, handoffs, ownership, and real-time status. Agents can claim Goals with attached Knowledge, and work can hand off between team members and AI agents. Needs attention surfaces Decisions that require human intervention. Humans Accept or Reject those decisions, keeping judgment calls explicit instead of buried in chat or agent logs.
4. Completion Evidence Packs and Documentation Write-Back: The Complete stage gathers structured proof for every finished Goal: test reports, screenshots, pull requests, and links. Optional human sign-off can be added as an extra review layer, but it is additive rather than a silent deployment gate. Every result traces to the originating Goal and Decision. Evidence and outcomes write back into the Knowledge base, so documentation reflects what actually shipped.
5. Idea Capture and Goal Shaping Tools: Rinkata includes a notebook drawer for quick notes. Notes can be promoted to Ideas or added to Knowledge. An agentic system can debate and refine ideas, auto-fill details from a single line, and add acceptance criteria from the beginning. Ideas can also be extracted from uploaded documents with one click. Documentation can be organized flexibly, giving teams a lightweight product discovery workflow before agent execution.
6. Agent-Agnostic CLI and Integration Layer: Rinkata installs once and connects to existing coding agents through the rinkata CLI and agent integrations. Supported agents include Codex, Claude, ChatGPT, Cursor, Grok, and Gemini. Teams keep their existing agents, GitHub workflow, and review habits. Hub owns Goals, specs, tickets, decisions, and evidence; local files act as cache or import/export surfaces, not a parallel source of truth agents should edit manually. Installation is available via curl -fsSL https://rinkata.dev/install.sh | bash.
7. Role-Based Workflows for AI-Native Teams: The same Rinkata Hub UI supports founders, product leads, engineering managers, and agent operators. Founders preserve north-star intent while multiple agents build in parallel. Product leads turn rough ideas into testable Goals with acceptance criteria. Engineering managers see blockers and judgment calls. Agent operators hand work between agents without losing context and attach evidence to each handoff.

Problems Solved

1. Pain Point: AI-assisted development often suffers from context fragmentation, intent drift, and stale documentation. Coding agents may invent missing context, miss specifications, duplicate work, or make decisions that humans never approved. Project knowledge lives across chats, documents, repositories, and issue trackers. Completed work lacks an audit trail connecting tests, screenshots, and pull requests back to product intent. Manual documentation falls behind, and handoffs between agents or humans lose rationale. Rinkata addresses these problems with shared memory, explicit Decisions, and completion evidence.
2. Target Audience: Rinkata is built for software product teams using AI coding agents, including founders, product managers, product leads, engineering managers, technical leads, developers, and agent operators. It fits teams adopting Claude, Codex, Cursor, ChatGPT, Gemini, or Grok in parallel, and organizations that need agent-readable specs, human-in-the-loop review, and traceable completion evidence. It is also relevant for teams that want a shared source of truth without replacing their repository, IDE, or coding agent.
3. Use Cases:
   • Multi-agent product development where several agents work in parallel and must stay aligned to the same Goals, specs, and decisions.
   • Product discovery handoff where a rough idea is captured, refined into a Goal with acceptance criteria, and assigned to an agent.
   • Engineering handoffs where the next agent receives covering Knowledge, prior decisions, and current progress without re-inventing context.
   • Compliance or quality-focused delivery where test reports, screenshots, pull requests, and links are collected as proof.
   • Documentation automation where completed work writes back into Knowledge, keeping docs current.
   • Human-in-the-loop review where Decisions surface under Needs attention and require Accept or Reject before proceeding.
   • Agent-agnostic orchestration where teams switch between Claude, Codex, Cursor, ChatGPT, Gemini, or Grok while keeping project truth in Rinkata.

Unique Advantages

1. Differentiation: Rinkata differs from standalone issue trackers, documentation platforms, and AI coding agents by acting as an alignment layer rather than a replacement. Issue trackers manage tasks but not MCP-readable agent context. Documentation tools store knowledge but rely on manual maintenance. Coding agents generate code but can drift from product intent. Rinkata connects Goals, specs, Decisions, and Evidence in one shared hub that both humans and agents read and write. It is agent-agnostic, keeps existing GitHub and review workflows, and does not modify the repository as a parallel source of truth.
2. Key Innovation: The core innovation is the combination of an Intent Thread, MCP-based shared memory, human judgment gates, and automatic evidence write-back. Agents search Knowledge before inventing context, humans make the judgment calls through Needs attention Decisions, and finished work ships with proof that writes itself back into docs. Graph artifacts such as Goals, specs, and tickets become machine-readable project truth. The rinkata CLI and integrations let Codex, Claude, Cursor, Grok, Gemini, and ChatGPT participate in the same product intent lifecycle.

Frequently Asked Questions (FAQ)

1. What is Rinkata? Rinkata is an agent-agnostic AI product alignment hub and shared memory layer. It keeps product intent connected from idea to deploy by storing Goals, specs, Decisions, and completion evidence in one source of truth that humans and AI agents can read and write over MCP. It is not a replacement for your repository or coding agent.
2. How does Rinkata connect to AI coding agents like Claude, Codex, Cursor, and Gemini? Rinkata connects through the rinkata CLI and agent integrations. Codex, Claude, Cursor, Grok, Gemini, and ChatGPT can read and write project truth in Rinkata Hub UI so they stay aligned with the same Goals, specs, and tickets. This uses MCP for shared context retrieval and write-back.
3. Does Rinkata write code or modify my repository? No. Rinkata does not write code for you; you connect your own coding agents. Hub owns Goals, specs, tickets, decisions, and evidence. Local files are cache or import/export surfaces, not a parallel source of truth agents should edit by hand. Rinkata is the alignment layer, not a repo replacement.
4. What counts as completion evidence in Rinkata? Completion evidence includes structured proof such as test reports, screenshots, pull requests, and links attached at the Complete stage. Optional human attestation can be added as an extra review layer, but it is additive and not a silent gate that makes finished work look unfinished. Evidence traces back to the originating Goal and Decision.
5. When does an agent need human approval in Rinkata? An agent needs human approval when judgment is required, especially for Decisions surfaced under Needs attention. Agents may propose options, but humans Accept or Reject in Rinkata Hub UI. This keeps critical path choices explicit, auditable, and aligned with product intent.