Built for solo founders wearing every hat and compliance officers who own the final gate. Every Claude output cites a mechanism, clears a 14-framework compliance check, and carries a confidence rubric you can hand to an auditor.
Free Starter tier, forever. 90-day Pro Trial auto-activates on install. 300 free 12-month licenses available through the Early Adopter Program.
Now on every Claude surface. Claude Code (100% fidelity) · Claude Desktop (~95% via 9 MCPs + one-command setup) · Claude.ai Web (~85% via project instructions) · Cowork (~85% via plugin).
/plugin marketplace add DrNabeelKhan/maxim /plugin install maxim@maxim-packs
Six gaps every operator hits with an off-the-shelf LLM. Each one is a task you probably do manually today. Each one is a task Maxim does automatically, on every output.
Generic Claude writes with the same confidence whether it's citing a fact or making one up. Maxim tags every output 🟢 / 🟡 / 🔴 with a four-line audit trail so you know exactly what is solid and what needs a second look.
Monday's landing page sounds nothing like Friday's follow-up email. Maxim loads your voice profile on every session and enforces it on every output, so your copy stays on-brand without you re-pasting guidelines.
You ship a spec, legal flags it, you rewrite. Maxim enforces 14 compliance frameworks (GDPR, HIPAA, PCI-DSS, SOC2, and 10 more) at generation time. Regulated outputs either clear the gate or get blocked before they leave the model.
Claude's context window forgets. You re-explain your project, your customer, your stack. Maxim carries architecture, decisions, and handoff state across sessions so your second chat is sharper than your first. Your tenth is sharper still.
Asking Claude to design a pitch deck, audit a privacy policy, and review a Go service in the same session gets you three mediocre outputs. Maxim routes each task to a specialist agent trained on its domain. You get CEO-level strategy, CTO-level architecture, CSO-level security review: from the same terminal.
A generic LLM says "here's a landing page." Maxim says "here's a landing page applying Fogg B=MAP as the primary composition, with Prospect Theory in the CTA and pre-attentive attribute theory in the visual hierarchy." You get the mechanism, so you can evaluate the work.
Six real tasks you probably ran this week. What a generic LLM gives you. What Maxim gives you. Decide for yourself which one ships.
You get a polished email. You have no idea if it applies any persuasion framework. You edit for brand voice by hand.
You get a polished email composed with Fogg B=MAP, Maxim's Prospect-Theory CTA, and your brand voice already loaded. Confidence tag + gaps flagged.
You get a general review. GDPR and PIPEDA may or may not come up. You chase the regulator for clarification.
Claude routes to CSO · security-analyst. 14 compliance frameworks load automatically. Every clause is tagged against the relevant regulation with cited section numbers.
Context window is empty. You re-paste your README, your sprint notes, your architecture doc. You re-explain who the customer is.
Session-continuity loads your project's decisions, open handoffs, and moat claims on start. Thirteen drift checks run automatically. You continue, not restart.
You find out in review. Or worse, in production. You rewrite the spec.
Proactive Watch catches the reference against your dependency manifest before the spec leaves the session. Drift class: orphan-refs. Severity: 4.
There's no rubric. You trust it or you don't.
Every output ends with Basis · Gap · Mitigation · Next. You know what grounds the answer, what's missing, how to close the gap, what would change the tag. Trust becomes mechanical.
Everyone has their own prompt library. Brand voice is a Google Doc nobody opens.
Three-layer voice system loads Maxim base, operator overlay, and startup overlay on every session. Your team's outputs sound like one team, not five freelancers.
Six reasons operators who use Claude every day install Maxim on day one. Each reason has a behavioral science framework behind it: footnoted, not shouted.
A failed compliance audit. A brand voice that slips in a prospect email. A product spec that hallucinates a library. A pricing page that contradicts the contract. These cost real money, real trust, and real time to unwind. $19.99/month is what most teams spend on coffee for a week.
Framework: Prospect Theory (Kahneman & Tversky, 1979) →Paste the install command. Hit Enter. Your very next output comes back with a confidence tag, a cited framework, and an audit trail. No retraining, no new workflow, no handoff between tools. You keep using Claude Code exactly the way you already do: the outputs just come out better.
Framework: COM-B (Michie, van Stralen & West, 2011) →Generic Claude hands you confident prose. Maxim hands you prose plus the behavioral mechanism it applied and the anti-patterns it avoided. You evaluate the output against the claim. You can defend it to your boss, your customer, or a regulator. It stops being a chat. It starts being documentation.
Read the ADR: Behavioral Moat Framing Doctrine (ADR-007) →Every Maxim response ends with a four-line rubric: what grounds this answer, what is missing, what you should do to close the gap, what would move the confidence up or down. You stop guessing when to trust the model. Trust becomes a property of the output, not a property of your mood.
Framework: Technical Educator Rubric (ADR-010) →Starter is free forever. Not crippleware. Not a 30-day trap. A regression test in our build fails the release if the free tier ever silently narrows. Start free. Upgrade when the moat shows up in your work. Cancel any time.
Read the ADR: Reciprocity (Cialdini, 2001) + ADR-004 →Claude's context window forgets. Maxim carries architecture, decisions, skill gaps, and handoff state across sessions. Thirteen drift checks run on every session start: docs vs code, counts vs reality, moat claims vs the ledger. Silent regressions surface before they ship, not after.
Read the ADR: Zeigarnik Effect + ADR-002 Executable Contracts →This section was composed through Maxim's own behavioral intelligence skills. Every block names the framework it applied. A tool that claims behavioral rigor should demonstrate it visibly.
A generic LLM wrapper gives you prompts. Maxim gives you mechanism, citation, and a registry of anti-patterns, applied automatically.
Every output cites a peer-reviewed behavioral framework with author and year. Framework. Mechanism. Anti-pattern. Reviewed in SKILL.md before a pack ships.
Fogg B=MAP · Cialdini · Prospect Theory: author + year cited
Confidence tags earn their color. 🟢 HIGH only when the skill matched, the framework fired, and the gap log is clean. 🟡 and 🔴 name the gap.
Confidence rubric · Basis · Gap · Mitigation · Next
14 frameworks wired into the MCP layer. GDPR, HIPAA, PCI-DSS, SOC2, and eleven more. Not a checklist: a blocking gate on outputs that touch regulated data.
14 frameworks · mxm-compliance MCP
Thirteen drift classes scanned on session start. Docs vs code. Counts vs filesystem. Moat claims vs MOAT_TRACKER. Silent regressions surface before they ship.
13 drift classes · every session start
Unattended workflows with a budget guard that hard-kills on token, tool-call, runtime, or cost breach. A verification gate runs separately from generation, and dry-run is on by default. Autonomy you can actually let run.
Budget guard · verification gate · dry-run default
Every prompt routes to the right executive office automatically. You never have to remember a command. The behavioral layer fires whether or not you ask, with the routing cost shown up front.
Routes every prompt · Claude Code CLI
Four tiers. Annual billing saves two months. Team ($249, 5 seats) and vertical overlays on the full pricing page.
The full governance substrate. Permanent, not crippleware.
The behavioral intelligence specialist. MOAT anchor.
Solo plus compliance depth and full watch coverage.
Pro plus unlimited brand / design and priority support.
A regression test fixture verifies the Starter feature set against filesystem reality on every build. The free tier does not quietly narrow. If a paid tier absorbs a Starter capability, the commit fails. If we ever change scope, it requires a visible ADR amendment, never a silent release note.
Starter stays free forever. In four years it becomes Apache 2.0 per BSL 1.1.
Generic Claude (and GPT, Gemini, and every LLM) gives you confident but ungrounded output. Maxim converts that into governed, citable, auditable work. Every output cites a behavioral framework with author and year. Every output carries a confidence tag with a four-line rubric. Every session runs drift detection across 10 classes. Every regulated workflow is blocked or allowed at the MCP layer based on 14 compliance frameworks. If you use an LLM for consequential work (compliance, brand, product, architecture, regulated data) you need enforcement, not advice. Maxim is that enforcement.
Prompt libraries give you templates. Chat wrappers give you UI. Maxim gives you a governed multi-agent operating system. 91 specialist agents with explicit dispatch through 7 executive offices. 52 skill domains where each SKILL.md names the mechanism it applies. A MOAT registry that prevents unbacked marketing claims. An Executable Contract protocol that fails commits when ledgers drift. An audit trail on every output. This is infrastructure, not prompts.
Starter is free forever, not crippleware, but the full governance substrate with 91 agents, 52 skills, 50 commands, and basic compliance advisory. A 90-day Pro Trial auto-activates on install. After that, Solo is $19.99/month (the MOAT anchor), Pro is $39, Professional is $99, Team is $249/month for 5 seats. Vertical overlays (Healthcare, Legal, Fintech, GovTech) are one-time purchases stackable on any paid tier. Annual billing saves two months.
BSL 1.1 is source-available, not open source. You can use, modify, fork, and redistribute Maxim's core framework for almost any purpose: internal tooling, client engagements, research, education, commercial services that integrate Maxim. The one restriction is that you cannot resell a competing pack catalog. Exactly four years after each version release, that version converts automatically to Apache 2.0. Maxim v1.0.0 (released 2026-04-21) becomes Apache 2.0 on 2030-04-21.
Maxim is built for Claude (Anthropic) via Claude API, Claude Pro, Claude Max, or Claude Code. The dispatch architecture is model-agnostic by construction (declared in config/project-manifest.json), with cross-provider routing to OpenAI, Gemini, Mistral, and local models planned for v8.0+. Maxim is an add-on to Claude, not a standalone product.
Fourteen frameworks wired at the MCP layer: GDPR, PIPEDA, PCI-DSS, SOC2, HIPAA, UAE-PDPL, CASL, FINTRAC, EU AI Act, ISO 27001, ISO 13485, ISO 14971, NIST CSF, and WCAG 2.1. Enforcement is blocking: outputs that touch regulated data without the required controls are prevented, not flagged. Vertical overlays add industry-specific depth for Healthcare (HIPAA + ISO 13485/14971 + FHIR/HL7), Legal (privilege + ethics walls), Fintech (PCI-DSS + FINTRAC + SOX + FDX), and GovTech (FedRAMP + NIST 800-53 + Section 508).
In Claude Code, Maxim installs via either the Anthropic official directory (/plugin install maxim@anthropic-official) or Maxim's own marketplace (/plugin marketplace add https://github.com/DrNabeelKhan/maxim followed by /plugin install maxim@maxim-packs). Either path installs 91 agents, 52 skills, 50 slash commands, 9 MCP servers, and 16 hook scripts. Commercial packs install from the Maxim marketplace after purchase, for example: /plugin install mxm-pack-l1-6-behavioral-intelligence@maxim-packs.
Operators who rely on LLMs for consequential work: growth marketers and SaaS PMs who need conversion copy to cite behavioral mechanisms, AI-savvy consultants and indie builders who need the full governance stack at individual pricing, compliance-conscious teams in regulated industries (healthcare, legal, fintech, govtech) who need enforcement not advice, and solo founders shipping products where the cost of a silent drift is higher than the cost of a $19.99 subscription.
90-day Pro Trial auto-activates on install. Keep using Starter after, or commit to Solo at $19.99.