An MD spec is a flat document. Memex makes it an enforceable database
It hosts the why, decisions, tasks and AC, forcing the agent to focus on each task with the exact slice of relevant context.
For frontier builders running AI coding agents at scale
Serious about AI-first development? MD docs are holding you back.
MD docs are why agents build things wrong & are not a living team wide context layer. Instead, spec features with your agent and push them into Memex via MCP. Each decision you make becomes a locked-in agent task with an AC it must build to, each spec compounding into a proactive, living brain for your team.
Works with your tools
- Slack
- Codex
- Copilot
- Cursor
- Claude Code
The Problem
MD docs and Issues were never built to give agents context at scale.
Never truly follows the spec
A doc is just prose, so nothing holds your agent to tasks. It interprets, skips work, and hallucinates.
Death by MD files
Hundreds of MD files, never all up to date. Neither human nor agent knows which to trust as the codebase changes.
Endless re-work & review
Endless re-prompting and reviewing to get it right. Worse on long-horizon tasks & brownfield code.
The solution
Memex AI is the spec-driven development system that makes your agent build it right first time, on a context layer that scales the more your team builds.
- Coordination & Intent layer: Jira / GitLab / Linear. Who is doing what, When does it ship, Roadmap & sprint planning, Manager dashboards. System of record for: Who + When + Status.
- Specify & Verify layer: Memex AI. What we're building & why, Decisions, first-class & queryable, Acceptance criteria, verifiable, Standards & drift detection, Knowledge graph across specs. System of record for: What + Why + Done-means.
- Implementation layer: Claude Code / Cursor / Copilot / Codex. The actual code, Reasoning, Access to your repo. System of record for: How + The actual change.
assigns & tracks work. agents query & build from.
How Memex AI fixes the problems of MD docs & Issues for AI coding
MD context mgmt systems are just outdated docs. Memex specs become a shared, living second brain that makes everyone faster.
Always current, no upkeep. Memex surfaces what teammates are deciding, catches when your work collides, and makes every spec faster than the last.
With an MD spec, "done" is just the agent's word. Memex proves it in CI.
Each task carries acceptance criteria, verified in CI, so "done" is proven, not claimed.
You edit an MD file alone. Memex specs let humans & agents collaborate live.
Humans and agents work one live spec you share, comment on and review or even push into Slack, not a hard to read MD file passed by hand.
Agents skim your MD codebase rules once. Memex enforces them on every task they claim.
Every task is checked against your rules. When code and a rule disagree, Memex flags it and you decide: correct the code, or update the rule.
How it works
Use Memex AI when you are ready to specify, build then verify a new feature with AI.
Specify
Paste or write the spec. Memex AI surfaces the critical decisions & conflicts you need to resolve.
- Paste an existing spec or doc, or push one straight from your coding agent conversation. Memex structures it into a living spec.
- Critical and blocking decisions extracted first, checked against prior decisions and active specs, routed to the right person, then turned into tasks for the agent.
- Build is blocked until every decision is resolved. Each resolution automatically becomes a scoped agent task.
Build
Your agents work through tasks in your IDE, logging every decision they make, escalating the ones they can't.
- Start your coding session, ask for tasks, and each agent gets a scoped context payload via MCP: the decisions and standards for that exact task. Not a static MD file.
- Mid-build decisions surface automatically: resolved against your standards where they're clear, escalated to a human where they're not. Every decision is logged back into the live spec in real time, and execution continues when resolved.
- Watch the spec board update live as agents work. Phase and assignment advance automatically as agents complete tasks, so the Kanban is never stale.
Verify
Every acceptance criterion verified automatically. Done is a fact, not a judgement call.
- Acceptance criteria are written into the live spec during planning. Agents link tests with a single comment line, no new tooling is needed. Status derives from CI: verified, failing, stale, or untested.
- The verify gate shows exactly which ACs are still untested before you call the spec done. Coverage is a fact on the board, not a guess.
- Bugs convert to agent work with a verifying AC attached and close themselves when the test goes green. And if a later change breaks an AC from this spec, CI catches it and flags it for a human.
Common questions
Does it replace Jira or Linear?
No. Trackers own who is doing what, when it ships, and status, the work before a spec is ready. Memex is the intent layer between your tracker and your code. Some teams find it enough on its own, but it's designed to work alongside Jira, GitLab and Linear.
Does it replace my coding agent?
No. Memex is a harness for the coding agents you already use, connected over MCP.
How is this different from CLAUDE.md, AGENTS.md or Cursor rules?
Keep them and run Memex alongside, or turn them into standards. A standard is a structured, enforced version of a CLAUDE.md: consulted on every task an agent claims and searched semantically, not a flat file skimmed once. When code and a standard diverge, the Drift Inbox flags it with a fix.
Isn't this spec-as-source or MDD all over again?
Spec-as-source, and its ancestor MDD (model-driven development), generate all your code from the spec one-to-one, with humans never touching it. It kept breaking because real systems are too messy for that. Memex is spec-anchored: your real code stays in your repo, kept honest by your tests, and we add the CI loop without generating code from the spec.
Do I have to change my workflow?
Keep planning in your tracker. Once a ticket is prioritised, you write one living spec in Memex. The tracker and your repo stay exactly as they are.
Won't the spec just become slop I can't keep in sync?
You don't keep it in sync by hand. Decisions and Issues are written back as the agent builds, so the spec is fed by the build, and any drift from your standards is flagged with a fix.
What actually lives in a spec?
Six linked records: the narrative (what and why), decisions, acceptance criteria, tasks, issues, and the test signals that verify it.
How does my coding agent actually use it?
Memex is MCP-native. Your agent lists your memexes, queries a task's scoped context, resolves decisions and writes decisions and Issues back, all over MCP, inside the same spec you see. One command wires it up.
What are standards?
Your codified rules and conventions. Every rule is a clause an agent must consult on every task it claims, injected into its context, not a style guide it forgets.
Make your agents build the thing right, first time
Get started freeNew to Memex? Download the Understanding Memex guide (PDF).