Your AI coding agents are brilliant.
But you don't know what they're building.

Memex AI is the strategy, decision, and knowledge layer for AI-native software teams. It gives every agent, and every human, the shared context that turns individually capable into collectively coherent.

memex-cli

The world moved on. The tools didn't.

Decisions are invisible in both directions

The old failure mode: human decisions buried in Slack. The new failure mode, growing every day: decisions locked in human–AI conversation threads nobody else will ever read. Two engineers, two agents, two contradictory choices made on the same Tuesday afternoon.

#eng-backend 3 weeks ago
"let's go with Valkey instead of Redis"
→ agent builds Redis. PR rejected.
AI planning session yesterday
decided: session tokens, not JWTs
→ locked in a thread. nobody else knows.

Knowledge decays

Conventions and procedures live in READMEs that were accurate six months ago. For AI coding agents, stale docs aren't friction. They're fatal.

testing-guide.md
last updated: 6 months ago
says: "mock the database"
→ team stopped after prod incident

Agents can't coordinate

Two agents work on related features simultaneously with no mechanism to share context. Neither knows the other exists.

Agent A: modifies auth interface
Agent B: builds against old interface
→ merge conflict. both agents' work wasted.

The bottleneck in modern software development is no longer writing code. It's making decisions, and making sure everyone knows what was decided, why, and what it means for the work ahead.

Built for the team you actually have.

Software teams are humans and AI coding agents now, working in parallel, often without seeing each other's work. Memex AI is designed for both from the ground up.

Humans set direction

Define strategies, make decisions, and articulate the why machines can't infer from code.

The shared graph

One source of truth that both humans and machines read and write to.

Agents execute with context

Agents act on what was actually decided, not what they hallucinate from stale docs.

What humans bring

  • Strategy and intent: the why behind the work
  • Decisions with rationale that others can trace
  • Priority judgement when trade-offs have no clear answer
  • Review of agent-extracted decisions and execution plans
  • Domain knowledge that doesn't exist in the codebase

What machines bring

  • Tireless execution against well-defined work items
  • Automatic extraction of decisions from conversations
  • Drift detection: flagging when docs diverge from code
  • Dependency checking before every implementation
  • Parallel work across the entire codebase, simultaneously

Not a human tool with an API bolted on. Not an AI tool with a dashboard bolted on.
One graph. Two interfaces. Zero drift.

You are not redundant. You are the chef.
AI is the sharpest knife you've ever held.

Our roles in software are evolving. Our humanity is not.

The grind is lifting. The tedium is collapsing. What's left is the part that was always the point: deciding what to build, why it matters, and whether it's any good.

  • You taste. You see the shape of the problem before anyone else, and you know when something is off even when every requirement is met.
  • You choose the dish. You weigh the trade-offs and pick the direction the team cooks in.
  • You run the pass. You own the recipe, the standard, the craft. Nothing leaves the kitchen without your say.
  • You hold the knife. AI is extraordinary at the cuts. It is not the cook. The dish is still yours.

Memex is built around that belief. Every primitive exists to keep the knives sharp and out of your way — so your attention goes to the part only the chef can do.

Five primitives. One graph.

Four primitives describe the work. One describes the people doing it. Together they form a single graph every human and every agent reads from and writes to.

Strategy: the objective
The container for everything.

A strategy articulates what the team is trying to achieve and why. Not a one-line epic title. The full context that makes every downstream choice intelligible.

Old world: Epic.  New world: a living document with reasoning, constraints, and context that agents can navigate.
Decisions: the leading edge
First-class objects, not afterthoughts buried in Slack.

Every non-obvious choice gets an ID, status, options, and rationale. Open decisions block. Resolved decisions constrain. When priorities shift, you re-resolve, and the system traces the impact.

D7: Cache strategy [OPEN]   D3: Auth approach [RESOLVED]
Work Items: the dependency graph
Not a backlog. A DAG.

Work items have goals, acceptance criteria, and explicit dependencies on other items and decisions. When scope expands, a work item gets promoted to its own strategy, with full lineage preserved.

WI-1 ← WI-2, WI-3  │  WI-3 ← D7, D8  │  WI-4 ← WI-1, WI-3
Blueprints: the shared contracts
Institutional knowledge that can't rot.

Scoped, prescriptive rules: not "here's how auth works" but "when you modify auth, do X, never do Y, verify with Z." Every rule is provenance-tracked — each one links back to the decision that produced it, so you can always see why the blueprint says what it says.

deployment [CURRENT]   testing [DRIFT DETECTED]
Humans: the creative drivers
First-class, named, and routable.

Humans aren't "whoever happens to be logged in" — they're an explicit primitive in the graph, connected to every other one. Memex watches who creates which strategies, who resolves which kinds of decisions, who writes which blueprints, whose reviews catch the issues that matter. Over time it builds a picture of who holds what, and routes the right decisions to the right people, automatically.

@sarah owns auth · @alex owns deployment · @priya owns design-system

A strategy is built where the code lives.

In the old world, planning happens in a separate universe from implementation. Someone writes a spec in a doc, shares it in Slack, it goes through a review cycle, and eventually an engineer reads a half-stale document and starts coding from an incomplete understanding of the intent.

In Memex, you build the strategy inside your coding tool, through conversation, grounded in the actual codebase. The agent that helps you draft it is the same agent that will pick up the first work item.

01

Problem framing

You articulate the problem. The agent reads relevant code, drafts the purpose statement. You push back. The purpose sharpens with each pass.

02

Decision surfacing

As the approach comes together, the agent recognises non-obvious choices. It creates draft decisions with options and trade-offs for you to resolve.

03

Work item scoping

The agent reads the codebase to understand what exists, what needs changing, and what the dependency order should be. Then proposes work items.

04

Blueprint linking

Existing blueprints are linked to relevant work items. New blueprints are scoped where gaps exist. Publish, and the graph goes live.

strategy-draft
you → I want to add proactive role discovery, help me think through this. # agent drafts the strategy, reads the codebase, surfaces decisions agent → create_strategy_draft("Proactive Role Discovery") agent → add_draft_decision("server-side vs client-side matching", [A, B]) agent → add_draft_work_item("Profile schema redesign") agent → add_draft_work_item("Discovery matching engine", deps=["WI-2", "D6"]) you → let me see where we are agent → get_strategy_draft("S3") memex ← # S3: Proactive Role Discovery [DRAFT] 3 decisions open, 5 work items scoped, 2 blueprints linked # when it's solid, publish you → looks right, publish it agent → publish_strategy("S3") memex ← strategy active · decisions blockable · work items claimable
No context switch. The strategy is built where the code lives.
No handoff. The agent that co-authored the strategy has full access when it implements.
Grounded in reality. The agent reads the codebase during planning. The strategy reflects what actually exists.
Multiplayer. The draft lives on the server. Teammates contribute concurrently from different environments.

The system works for you.

Memex AI doesn't just store knowledge. It actively detects when things go wrong and surfaces decisions you'd otherwise miss.

The execution plan gate

The single most effective quality gate for AI coding agents.

  • Reconcile first: before any code is written, the agent produces a plan grounding the spec in the actual codebase
  • Surface conflicts: the plan lists files to modify, dependency flow, and where the design contradicts reality
  • Human review: no coding happens until the plan is reviewed
  • Stops hallucinations cold: the agent can't drift into an imagined implementation
plan queued approved → implementation

Drift detection

The feature that kills the wiki.

  • Agent-reported: agents flag blueprint/code mismatches during execution planning
  • Decision-triggered: resolved decisions auto-mark affected blueprints for review
  • Implementation-triggered: completed work items prompt blueprint review
  • Scheduled audits: continuous automated comparison, not quarterly
deployment [CURRENT] testing [DRIFT DETECTED]

Decision extraction

Nobody will log them manually.

  • Passive extraction: decisions surfaced automatically when sessions end
  • Decision bundles: batched with summary, rationale, and source link
  • Review like a diff: approve, reject, or flag with progressive disclosure
  • Merge into the graph: approved decisions become first-class objects
3 decisions extracted → 2 approved, 1 flagged

MCP-native. Connect any agent.

Memex AI is an MCP server. Any AI agent connects to the same decision graph through standard tool calls — your coding agent, a research agent investigating open decisions, a testing agent verifying completed work, a custom agent you built this afternoon. One protocol, one graph, every agent.

mcp-tools
# Strategy (top-level container) list_strategies() all strategies with status summary get_strategy(id) purpose, vision, decisions, work items promote_work_item(wi_id) elevate a WI to its own strategy, preserving lineage # Strategy drafting (collaborative design in your coding tool) create_strategy_draft(purpose) start a new strategy in draft state add_draft_decision(id, q, options) surface a design choice during planning add_draft_work_item(id, goal, deps) scope a unit of work with dependencies get_strategy_draft(id) current state, formatted for reading publish_strategy(id) draft → active · decisions become blockable # Decisions (within a strategy) get_decision(id) status, options, rationale resolve_decision(id, choice, why) resolve + cascade impact reopen_decision(id, reason) re-open, cascade staleness get_decision_impact(id) WIs blocked, blueprints affected, cross-strategy deps # Work Items (dependency graph) get_ready_work_items(strategy?) what can start right now submit_execution_plan(wi, plan) ground truth before coding check_dependencies(wi_id) met vs blocked, within + across strategies # Blueprints (cross-cutting knowledge) get_blueprint(domain) scoped context for agents flag_blueprint_drift(id, evidence) "code says X, blueprint says Y" get_blueprints_for_work_item(wi) which blueprints to load # Decision Extraction extract_decisions(session_id) surface decisions from conversations create_decision_bundle(decisions) batch for review, like a merge request get_pending_bundles(account_id?) bundles awaiting review approve_bundle(id) approve all decisions in the bundle review_bundle(id, actions) approve / reject / flag per decision

Works with Claude Code · Cursor · GitHub Copilot · Windsurf · Custom agents · any MCP-compatible tool

One MCP server. Every AI coding tool.

Add Memex AI to your agent's MCP config and the entire decision graph is available. No plugins, no vendor lock-in.

Claude Code
Anthropic
Native MCP support via claude_desktop_config.json
Cursor
Anysphere
MCP servers in Cursor settings
GitHub Copilot
GitHub
MCP support in Copilot Chat & agent mode
Windsurf
Codeium
MCP server configuration in Cascade
Cline
Open Source
Built-in MCP server management
Custom Agents
Any framework
Standard MCP client SDK
agent-workflow
# Agent picks up work via MCP agent → get_ready_work_items(strategy="S3") memex ← WI-2: Profile schema redesign [ready] # Agent loads context before coding agent → get_work_item("WI-2") agent → get_blueprints_for_work_item("WI-2") memex ← blueprints: [data-model, testing] agent → check_dependencies("WI-2") memex ← all dependencies met, 0 blocking decisions # Agent submits plan for review agent → submit_execution_plan("WI-2", plan) memex ← plan queued for review # After approval: agent implements, then reports agent → update_work_item_status("WI-2", "complete") memex ← WI-2 complete → WI-4 unblocked

Familiar tools. Different era.

vs. Jira / Linear / Shortcut

These track who's doing what. Memex AI tracks what was decided, why, and what agents need to know before touching the code.

vs. Confluence / Notion

Wikis have no drift detection, no provenance, and no scope boundaries. Memex AI blueprints are living documents with automated freshness guarantees.

vs. CLAUDE.md / cursor rules

These validate the need, but they're per-repo, per-tool, and manually maintained. Memex AI is the managed, multi-agent, cross-repository evolution with decisions, dependency graphs, and drift detection.

Three steps. No migration.

01

Start with the docs you already have

Point Memex at your Notion, your repo, or your Confluence. It reads what's there, drafts your first strategy, and surfaces the decisions already buried in your existing docs. Ready in minutes, not a migration.

02

Connect your coding agent

Any MCP-compatible agent — Claude Code, Cursor, Copilot, Windsurf — plugs in with a few lines of config. From the next session, it reads the graph before writing code.

03

Watch the cycle work

Agents load context, check decisions, submit execution plans for review. You approve, they implement. Drift detection starts day one.

Your Memex holds your most sensitive work.

Strategy, architectural decisions, security processes, unreleased product thinking, internal blueprints — all connected in one graph. That's the point. It also means self-hosting can't be an enterprise checkbox. It's a core design principle.

Hosted SaaS

We run it for you

Sign up, connect an agent, start building. The simplest path to getting your team onto Memex. Best for teams who want zero operational footprint.

Self-hosted

100% of your data stays inside your infrastructure

Memex ships as a single Docker container against any Postgres 15+ database. No Kubernetes cluster. No service mesh. No outbound calls for your graph, your decisions, or your blueprints.

One container, one Postgres Pull the image, point it at a database, done. A modest VM is enough.
No outbound dependencies Self-hosted mode makes zero calls to Mindset infrastructure for any core feature.
Bring your own LLM keys Anthropic, OpenAI, Google, a private model in your VPC — whatever your team has standardised on. Memex is provider-agnostic and bills to your account.
Portable by design Start on SaaS, migrate to self-hosted later, or vice versa. The graph is just Postgres.

If your codebase lives in your infrastructure, your Memex should be able to live there too. We've built the product so that's never a trade-off against functionality.

Simple, transparent pricing.

Cloud

$50/user/month

Managed by us. Start building in minutes.

Join the waitlist
  • Unlimited strategies, decisions & work items
  • Blueprint authoring & drift detection
  • Decision extraction from agent sessions
  • Full MCP server for all compatible agents
  • Dependency graph & impact analysis
  • Multi-agent coordination
  • Team collaboration & permissions
  • Cloud CDN & automatic backups
  • Email & chat support

Self-Hosted

Custom

Your infrastructure. Your data. Your LLM keys.

Contact Sales
  • Everything in Cloud, plus:
  • Deploy on your own infrastructure
  • Complete data sovereignty
  • Use your own LLM API keys
  • SSO & enterprise authentication
  • Priority support & onboarding
Requires Docker + PostgreSQL 15+

Software tooling always follows the bottleneck.

AI coding agents made implementation fast and cheap. The new bottleneck is strategy and decisions: articulating what we're building, resolving ambiguity, and ensuring every agent and every human acts on the same understanding.

Memex AI is the tool for this new bottleneck.

We built Memex because we lived the problem.

Mindset AI is the agentic platform powering in-app AI agents at a growing list of B2B SaaS companies. Every engineer on our team works with coding agents daily. Our product people design in conversation with AI. Our support team uses AI to triage and respond. We aren't just building AI agents — we're running an AI-native company.

The bottleneck shifted

When we started, we assumed the hardest part of AI-assisted engineering would be code quality. It wasn't. Modern coding agents produce good code when they know what they're trying to build.

The hard part turned out to be context. Two engineers, each with their own agent, arriving at contradictory decisions. An agent loading a wiki that was last updated two quarters ago. A re-prioritisation that silently invalidates half the decisions that shaped the previous plan.

We run Mindset on a set of mental models that take bottlenecks seriously. One of them is the Theorem of Constraints: any interconnected process operates at the speed of its slowest part. We're allergic to bottlenecks in our own organisation, and we look for them constantly.

When we applied that lens to AI-assisted development, the bottleneck was unmistakable: it wasn't writing code, it was capturing, propagating, and maintaining decisions across humans and agents.

Memex AI is the system we built to remove that bottleneck. It started as internal infrastructure. We use it every day. It's now the decision layer behind everything Mindset builds — and Memex itself runs on the Mindset agentic platform, so the product you're looking at is a working example of what both can do together.

Build with shared context.

Memex AI is in early access for teams already building with AI coding agents.

Because the hardest part of building software was never writing the code.