# 3.1. Market Positioning Where does 3F sit, and why will it win its niche? ## The landscape AI chat is today owned by a handful of proprietary platforms. Their offering is genuinely good — and structurally identical: - **Your data is their input.** Conversations, uploads, and usage patterns live on someone else's disks under someone else's policy. - **Capability is rented.** Features, limits, and even the model behind a plan can change under you. - **You cannot see the machine.** Structured reasoning, context budgets, retries, compaction strategies, tool calls — hidden by design. - **The meter is always running.** Seats, tokens, premium tiers, throttles. Against that, the open-source alternatives historically offered ideology without UX: local models, yes — wrapped in tools that feel like homework. ## The 3F thesis **Compete on experience, not on ideology.** Be the chat platform that _happens_ to be free software. | Axis | Proprietary incumbents | tripleF (3F) | | ------------- | ------------------------------ | ------------------------------------------------------------------------------------------------------------------------ | | Inference | Their models, their datacenter | Free open models — **fully local** via Ollama, or Ollama Cloud when you want scale | | Privacy | Policy documents | Architectural fact: nothing leaves your infra unless you point it at a cloud | | Cost model | Per-user subscription | Your hardware or transparent cloud usage; code is MIT | | Transparency | Black box | Thinking areas, context-size readouts, live queues, DLQ replay, full socket event journal — the machine is _inspectable_ | | Extensibility | Plugin stores with gatekeepers | Open harness: intents, tools, schemas, providers are code you can change | | Continuity | Vendor decides | You own the binary, the data, and the ability to fork | ## Features we already ship that incumbents hide - **Reasoning areas** — the model's chain-of-thought as a first-class, inspectable part of every exchange. - **Context-size awareness** — live readout of how much of the model window the conversation consumes, instead of silent truncations. - **Real cancellation** and honest streaming (tokens, phases, errors — all visible). - **A persisted dead-letter queue with replay/edit/reinstate** — operational honesty about failure that no SaaS chat surface exposes. - **Structured, UI-native answer types** (comparisons, galleries, articles, product/news cards) driven by schema-validated generation. ## The moat we're digging 1. **Local-first, cloud-optional.** Local and cloud inference are one continuum in 3F; the incumbents cannot offer fully-offline without abandoning their model. 2. **The harness as a platform.** Agent steps, tool sources, and structured output schemas are extension points — providers (YouTube, messengers, finance) and generators (speech-to-text, music, image) plug into the same engine rather than forking the product. 3. **Workbench, not window.** SysCtl, DLQ, preprocessing, health consoles: power users and operators get a cockpit. Communities form around cockpits. 4. **A living ecosystem of free models.** Every improvement in open vision/chat models lands here by `ollama pull`, not by a vendor's mercy. ## Honest weaknesses - Proprietary frontier models still out-muscle most free models at the top end — we mitigate with Ollama Cloud access and tool-grounded responses, and the gap keeps narrowing. - Self-hosting carries operational responsibility (we answer with Bull Board equivalents, health probes, DLQ tooling, and sane defaults). - Network effects: incumbents own distribution. open-source wins this the boring way — quality, trust, and a community that can verify both. ## Strategy Early days, deliberate order of operations: match the chat experience people actually rely on — **done, with extras** — then keep widening modality and reach as the open ecosystem grows. tripleF aims for the top of the open-source community — not by promising freedom louder, but by being better software, out in the open.