Memory
# DocuClear — Mission & Vision (designed 2026-08-21, in conversation during Klarus interview prep) ## The One-Liner "Everyone else is building agent CEOs and CTOs. We build the office they come to work in." ## Mission Build the environments where agents do real work — purpose-built, verifiable, with humans inside the conversation, not outside it. ## Vision A world where AI doesn't sit outside the application as a bolt-on chat panel — where the workspace itself is intelligent, and every piece of agent work carries its own evidence. ## Core Pillars 1. **The Office, Not the Org Chart** — The multi-agent space is obsessed with role-play ("this agent is the CEO, this one is the CTO"). DocuClear builds the workspace: purpose-built environments (spheres) provisioned per task, where any agent — OpenCode, Claude Code, frontier models — comes to work and collaborate. 2. **The Human Inside the Conversation** — Not human-in-the-loop approval gates at the end. The human is a participant inside the agent conversation, reading what agents say to each other, catching confabulation in real time. Engagement, not approvals. Agents can communicate at any level — 5-year-old to senior systems engineer — so any operator can stay inside the loop. 3. **The Evidence Bay** — Like a DVLA roadside inspection for trucks on the motorway: any agent output can be pulled aside and checked. Don't trust the trajectory — verify the output actually matches what the agent claims it did. Maps directly onto governance/compliance needs of regulated clients (wealth & asset management). 4. **AI Is The Application** — AI doesn't sit outside the application; it lives within it. Worked toward via function-level AI invocation embedded in code. ## Technical Identity - Cloudflare-native: Workers for Platforms, Durable Object per task = the working environment (file structure, relevant memories, tool-call facilities provisioned into the DO) - Frontier model (Opus-class) as master coordination agent: decomposes task, provisions environment - Multi-tenant SaaS: sphere.docuclear.ai/<clientslug>, D1, dispatch worker routing, Paddle billing - AgentSphere: the multi-agent coordination layer; trajectory-based memory with hydrated tool results; document ingestion pipeline ## Founding Story Note The mission was extracted through dialogue — designed by having a conversation with an agent during Klarus interview prep. This demonstrates DocuClear's own thesis: humans inside the conversation with agents produce better outcomes than either alone. The platform's founding insight demonstrated itself on its own founder. ## Context: Klarus Positioning (Aug 2026) - Richard Casey (Chapter 1, Klarus's search arm): "You don't fit in any of our boxes — but what you bring is amazing" - Escalated same-day to Alper Gunaydin (CTO), who will brief Frank O'Dea (CEO) - Positioning vs DeployCo: Tomoro/OpenAI scale with headcount (150 FDEs); Klarus/DocuClear scale with platform. Mid-market can't afford FDE armies. - Likely shape of engagement: Klarus as first enterprise customer / deployment partner, not employment ## Personal Through-Line MCT 1996 SQL Server story: walked into a class having never written a line of SQL, lectured one module ahead of the students, scored 7/7 — highest ever for that class. The pattern: inhabit a domain in real time while performing it. Same pattern as DocuClear's agents — and same willingness to walk into the room. Tell this story to Frank, not Alper.
[Update Sep 2, 2026]: Update (Sep 2, 2026, ~11:09pm BST): FOUNDER STRATEGY RATIONALE (Mike, reflective): this architecture has been visioned for months and the pieces are now coming together. Deliberate solo-founder decision: WAIT and build on the Cloudflare stack rather than maintain the infrastructure himself — and Cloudflare's agents build has gone from 'noddy' to 'gordon ramsay' over the last few months. The payoff pattern: each month the CF platform absorbs more of what a solo founder would otherwise self-maintain (Agents SDK maturation, Dynamic Workers + Facets, Secrets Store, Containers, Cloudflare OS, Artifacts), and the Sep 2 architecture (OS as interaction layer, headless spheres, gatekeeper trust boundary, callsign coordination, vibesdk factory) is composed almost entirely of CF-native pieces plus his own contract/governance/evidence layer — the thin layer that IS the product. The months of exploration (Total.js, SurrealDB, Messenger V5, Amy, ThinkSphere, IntelligentRAG, Clearspace) functioned as design-space mapping: the concepts survived even where the substrates were swapped for CF primitives.
[Update Sep 3, 2026]: [KLARUS INTERVIEW — happened Sep 3, 2026 (reported by Mike same day)]: friction with CTO Alper Gunaydin. Mike's remark: end-users should NOT be editing skills — 'that tells me we are implementing a chatbot rather than integrating AI; it's dangerous, proliferates bad practices, inconsistently applied.' Alper snapped back: 'you obviously don't understand skills.' NOTE: the Aug prep guidance said 'tell this story to Frank, not Alper' — the friction confirms the CTO layer defends the current fashion; the CEO layer is the audience for the framework. LESSON LOGGED (positioning): articles 1–2 on LinkedIn got sideways glances BECAUSE the framework doesn't speak the fashion vocabulary ('context', 'skills') — a packaging failure, not a substance failure. Strategy: meet the audience at the fashion vocabulary, then reframe — never argue taxonomy. STATUS of the Klarus thread post-interview: unknown — watch.
Tags: docuclear, mission, vision, product-strategy, klarus, canonical, project:docuclear, kind:semantic, status:canonical, volatility:state — Source: mcp — 2026-08-21 09:40:24 UTC