Product Vision

Canonical source: docs/01-executive-context/product-vision.md — this page is rendered from it on every publish; edit the source.

Product Vision

Canonical home for ClawAI's positioning — slogan, description, pillars, users and non-goals. Everything else links here: the README, the wiki, CLAUDE.md and every agent router. Decided in ADR-126 (2026-09-26); the change of direction is DRIFT-001. The flagship list, each traced to code, is flagship-features.md.

Every AI, one workspace.

Every frontier AI model in one workspace that sees, hears, researches and builds. Pay as you go, bring your team, or run it on your own hardware.

Replaced 2026-09-26: "Every Frontier AI Model, One Subscription".

---

Mission Statement

Put every capable AI model — cloud and local — behind one workspace where the work actually happens: talk to it, show it files, audio and video, let it research the web and write the documents, and pay only for what you use.

---

The Four Pillars

Pillar	Promise	What makes it true
One workspace	Not a chat box: the AI sees images and video, hears voice notes, reads archives, researches and crawls the web, compares and orchestrates models, and writes PDF, Word, Excel, PowerPoint and Zip files	Flagships 1–8 and 10
Pay as you go	Connector credit metered per use across every surface, shown in the visitor's own currency	Flagship 9; numbers in docs/business/
Bring your team	Admin-managed users, roles and permissions, plan grants, per-user usage, and coding agents an organisation can govern	Flagships 11–12, with the gaps listed there — no SSO and no shared team billing yet
Local-first and private	Run the whole stack on your own hardware, with Ollama, llama.cpp and local image models; virus-scanned uploads; retention and memory controls	Flagships 13–15

---

Problem Being Solved

1. One model is never enough, and switching is manual. Each provider is strongest at something different; people keep several subscriptions and copy between tabs.
2. Chat is too small for the work. Real tasks arrive as voice notes, videos, zip files and links, and end as a spreadsheet, a deck or a PDF. Most assistants handle one end of that.
3. Subscriptions charge for capacity, not use. A light month costs the same as a heavy one, per seat, per provider.
4. Privacy is all-or-nothing. Either everything goes to a cloud API, or you run a local model with none of the tooling around it.
5. Teams have no control plane. No shared user management, usage view or audit trail across the AI tools people already use.

---

Target Users

User	Primary need
Individual professionals	The best model for each task, files in and files out, paying only for what they use
Teams and business customers	Admin-managed users and roles, usage visibility, plan grants, governed coding agents
Developers	A coding agent in VS Code, research that reads the docs for them, every provider behind one key
Privacy-conscious organisations	The whole stack on their own hardware, local models, verifiable routing
Administrators and operators	Status, uptime, logs and metrics without shell access to production

Personas in detail: user-personas.md.

---

Value Proposition

For a team or organisation

• Every provider behind one workspace — 22 usable providers, cloud and local, routed by task, modality and context window.
• Pay for use — metered credit instead of a seat per provider.
• Control — users, roles, permissions, plan grants, per-user statistics, audit trail, and an admin status page with Grafana and Prometheus behind it.
• Your hardware if you want it — one installer, GPU overlay chosen for you.

For the individual

• It hears and sees — voice and video notes, images, PDFs, archives.
• It researches — an AI plans the web work, narrates each step, opens every link you paste, and crawls a site when asked.
• It builds — files written by the AI, named by the AI, in the format asked for.
• It remembers — memory, context packs and cross-thread context.
• Power tools — compare models with a judge, branch, edit and rerun, find in a conversation, read aloud.

---

Product Goals

1. The workspace is the product. Every orchestration mode takes files and research; every model gets media it can use, or a helper that describes it.
2. Every AI surface is metered, and every meter is honest. Integer money, reservation before the call, settlement after (rules/28, rules/37).
3. Routing picks a model that can actually do the task — right modality, prompt fits the window, provider not out of credit.
4. Self-hosted and portable. The whole platform runs on one machine with Docker Compose; cloud providers are optional.
5. Full operational visibility — health, uptime, logs, metrics, audit.
6. Localised in 13 languages, including right-to-left.

Non-Goals

1. Training or fine-tuning models. ClawAI orchestrates existing models.
2. Replacing each provider's native product. ClawAI is one workspace across providers, not a clone of any single one.
3. Real-time collaborative editing of a conversation.

Removed 2026-09-26 because they are no longer true: "does not browse the web or take actions" (it researches, crawls and runs a coding agent) and "conversations are single-user, no sharing" (public share links exist). Multi-tenant team accounts are not a non-goal — they are an open requirement (REQ-POS-005).

---

Success Metrics / KPIs

Metric	Target	How measured
Routing accuracy	≥ 85% positive feedback on AUTO-routed messages	ChatMessage.feedback aggregation
System availability	99.5%	Status page (Prometheus-backed uptime)
Metering completeness	100% of paid AI calls reserve and settle credit	PAYG ledger vs provider calls
Audit completeness	100% of AI interactions have audit records	Audit log count vs message count
Localisation coverage	100% of user-facing text in all 13 locales	i18n key coverage

Unknown - ask the owner and record the answer. — the commercial KPIs for the repositioning (pay-as-you-go conversion, team-customer count, self-hosted installs) have no targets yet.

---

Competitive Positioning

Dimension	Typical AI assistant	ClawAI
Models	One provider	22 usable providers, cloud and local
Inputs	Text and some files	Voice, video, images, PDFs, archives, links
Outputs	Text	Text, images, PDF, Word, Excel, PowerPoint, Zip
Web	Search snippets	AI-planned research, narrated, robots-respecting crawl
Pricing	Subscription per seat	Pay as you go, shown in your currency
Deployment	Their cloud	Hosted, or the whole stack on your hardware
Control	Basic	Roles, permissions, plan grants, usage, status page

Market detail: competitive-analysis.md.