Make your agents build exactly what you decided, not cowboy vibe-coding

The living spec your agents build from and verify against until your AC passes - not static MD docs outdated by day 2. Connected to your coding agents, every decision by people & agents throughout the build feeds back into your spec, automatically - so your agents build exactly what you decided.

AI coding agents are fast. But they build things you didn’t ask for. The more agents, the worse it gets.

Jira, Google Docs, MD files & Confluence were not built for continuous, automatic agent context management.

Slack decisions, codebase changes, agent calls mid-session - none of it updates what your agents build from.

Multiple agents, same feature, each making their own calls. Static MD docs as the only shared context.

Get early access

No credit card required.

Every decision before and during the build captured in the live spec, so everyone works from the same context. Always.

Before build, the implicit decisions hiding in your spec surface as blocking questions and route to the right person automatically. During build, every blocking decision an agent makes lands in the spec as a candidate for you to approve or reject, in the UI or from your coding agent. One live decision record every person and agent works from. No agent quietly decides your architecture for you.

A spec in build with three agent tasks — payment routes complete, auth middleware and webhook in progress — each logging the decisions it made.

Unlike CLAUDE.md and skills files agents skim once, your standards are applied and checked automatically on every task agents claim.

Unlike MD files and skills, every rule across your Memex standards is a clause agents must consult on every task they claim, injected into their context automatically. When an agent sees your code and a rule diverge, it flags drift; the divergence queues in your Drift Inbox with a diffed proposal to fix the code or change the rule. You stay in true control of your codebase.

A drift alert — standard-9 violated: retry logic uses 5 attempts but the standard allows 3 — offering to fix the code or update the standard.

Multiple agents. One decision graph. No passing context by hand with MD docs.

Instead of humans passing context around in MD files, every agent reads from and writes back to the same shared decision graph. Agent A’s decisions are immediately visible to agents B, C and D. When an agent hits something it can’t do, or finds a bug, it escalates to a human with its full reasoning captured, so you decide what happens next.

A shared decision graph around Spec 104 connecting agents A, B, C and D to their decisions.

Done is when every acceptance criterion passes in your CI, not when the agent says it’s done.

Acceptance criteria live in the spec. Agents tag a test with one comment line, and every CI run reports back: each AC shows verified, failing, stale or untested, live on the board. Months later, if any change breaks an AC from this spec, even someone working on something else entirely, CI flags it. Nobody has to remember to re-check.

Common questions

How is this different from CLAUDE.md or cursor rules?

CLAUDE.md and cursor rules only reflect what someone last edited. They live in flat files, accumulate contradictions, and can’t capture decisions made outside the repo. Memex AI is what that stack was trying to be - structured, stored in SQL, maintained automatically from the build itself, and delivered scoped to each agent per task.

How is this different from SDD tools like Kiro or BMAD?

Spec-driven tools generate a spec and hand it off. Memex keeps the spec alive through the whole build - every decision, whether made in chat with the agent or during review, writes back to the spec - and verifies the result against acceptance criteria before a PR. The spec is the source of truth at the end, not just the start.

Does this sit alongside Jira or replace it?

It sits alongside. Jira and Linear track who’s doing what; Memex tracks what was decided, why, and what an agent needs to know before touching the code. It’s the decision and knowledge substrate underneath your tracker, not a replacement for it.

How does it connect to coding agents?

Memex is an MCP server. Claude Code, Cursor, Copilot, Codex, and any MCP-compatible agent connect via standard tool calls and read from / write back to the shared graph before writing code.

What do I bring to start?

A repo and a spec, or just a conversation. Paste an existing doc or a conversation with your agent and Memex structures it into a spec. Six starter Standards are seeded into every new workspace.

Is there a free tier?

Yes, free for individual developers at launch. Team and self-hosted plans add multi-agent coordination, shared standards, and collaboration.

Your agents build from what your team actually decided.

Write the spec in Memex AI. Every decision captured in the live spec, every agent working from current context, every PR verified against what was agreed.

Get early access

No credit card required.

Make your agents build exactly what you decided, not cowboy vibe-coding

AI coding agents are fast. But they build things you didn’t ask for. The more agents, the worse it gets.

With Memex AI, your agents always build exactly what you decided.

Every decision before and during the build captured in the live spec, so everyone works from the same context. Always.

Unlike CLAUDE.md and skills files agents skim once, your standards are applied and checked automatically on every task agents claim.

Multiple agents. One decision graph. No passing context by hand with MD docs.

Done is when every acceptance criterion passes in your CI, not when the agent says it’s done.

Specification to verification - the full loop.

Paste or write the spec. Memex AI surfaces the critical decisions & conflicts you need to resolve.

Agents work through tasks - logging every decision they make, escalating the ones they can’t.

Every acceptance criterion verified automatically. Done is a fact, not a judgement call.

Start with your next feature.

Common questions

Your agents build from what your team actually decided.

Make your agents build exactly what you decided, not cowboy vibe-coding

AI coding agents are fast. But they build things you didn’t ask for. The more agents, the worse it gets.

With Memex AI, your agents always build exactly what you decided.

Every decision before and during the build captured in the live spec, so everyone works from the same context. Always.

Unlike CLAUDE.md and skills files agents skim once, your standards are applied and checked automatically on every task agents claim.

Multiple agents. One decision graph. No passing context by hand with MD docs.

Done is when every acceptance criterion passes in your CI, not when the agent says it’s done.

Specification to verification - the full loop.

Paste or write the spec. Memex AI surfaces the critical decisions & conflicts you need to resolve.

Agents work through tasks - logging every decision they make, escalating the ones they can’t.

Every acceptance criterion verified automatically. Done is a fact, not a judgement call.

Start with your next feature.

Common questions

Your agents build from what your team actually decided.

Get early access

You’re on the list.