Questions
What problem does Meridians solve?
Information is abundant; maintaining a trustworthy read of a moving world is still work. Feeds report activity, chatbots answer and forget, and notes go stale. Meridians keeps an evidence-backed model of the domain current, shows the few changes that alter its understanding, and turns that read into a reasoned next move toward your goal.
What is it, plainly?
A maintained AI domain expert you tutor and own. It runs one Program — Watch, Opinion, Tutor, Route — over a structured model of actors, rules, evidence, and open questions. Hosted keeps that expert working when every client is closed; Stories gives you the focused daily loop; the Studio opens the same model in depth. It is not a chatbot because the durable model and Route, not the conversation, are the product.
How do I use it day-to-day?
Return to one clear continuation. If something materially moved, Stories shows what changed, why it matters, and whether the expert needs your judgment. Inspect, accept, correct, or defer; confirm a Route move only when you are ready to take it. If nothing needs you, Meridians says you are caught up and shows the next run. Open the Studio only when you want to investigate or tutor the model deeply.
What's a Route?
A live, hedged plan toward a goal. Meridians first builds a causal Graph, then a Reading of the variable possibilities and timing, then compiles an executable decision graph of Stages, main line, complications, hedges, and monitors. You hold one current Position. Evidence may stage an advance or re-route, but only you confirm the move. The Route remembers the journey actually walked and is never scored right or wrong.
How does it decide what to believe?
It weighs evidence rather than treating every link alike: relevance to the open question, source quality, corroboration, recency, and the source's earned history all matter. Strong evidence moves a position; thin or conflicting evidence stays uncertain. Every material movement keeps its sources and before/after state so you can inspect what changed and correct the reasoning.
How is it mine, and why does it get better over time?
Your corrections become attributed, durable model changes instead of disappearing with a chat. Teach it which questions, sources, causal links, and distinctions matter; later research and Routes show the consequence. Months of evidence, corrections, and journey history compound into an expert another person cannot reproduce with one prompt. The canonical record remains readable and exportable even if billing stops.
Can I share my expert, or learn from someone else's?
The private product comes first. You can export your record and use bundled first-party starters, but a full Marketplace is deliberately later. It earns priority only after retained users love their experts, maintain them, and ask to share or discover others. When it arrives, imports are private forks with visible lineage and reviewable upstream changes; no update may overwrite your tutoring silently.
What does it cost?
Hosted is the recommended product at $5.99/week: one isolated, always-on Meridians runtime provisioned for you. License is $3.99/week and runs the same engine and record on your own computer. Both are Director BYOK, so provider usage is charged directly by your chosen providers and Meridians never marks it up. They are separate subscriptions; choose the host that fits you.
Can I make money from my expert?
Not in the first product release. Creator subscriptions and revenue share are a later layer, after a free exchange proves that users maintain experts others genuinely want. Meridians must first be valuable for one Director with one private expert; commerce cannot substitute for retention or product quality.
Is my data private?
Data sovereignty is a product boundary on both runtimes. Hosted stores the canonical record on your isolated instance volume and stores provider keys encrypted under the instance's data-root key; the Gateway never receives plaintext provider credentials or a second writable domain record. License keeps both on your computer. Billing may stop active work, but it never removes your ability to read and export the record.
Do I need a team to use it?
No. Meridians must deliver full value to one Director. You can later invite viewers, contributors, or managers into the same instance with explicit permissions, but collaboration is optional and never a requirement for activation.
Which AI does it use?
Meridians orchestrates frontier models through provider keys you supply. Those models perform bounded generation, research synthesis, and reasoning tasks; Meridians supplies the durable domain model, Program, permissions, provenance, and one-writer record that make the work compound. Better and cheaper frontier models improve the product without replacing its maintained reasoning layer.
What's the architecture underneath?
Each domain is a bounded, self-updating artifact with the same contract: System holds rules and constraints, World holds actors and state, and Fate holds open questions and their bearing. A reducer is the single state writer; the on-disk record is canonical; UI, Pilot, automation, and MCP use the same queries, actions, and operations. Separate domains compose through constellations without collapsing into one untraceable model.
How is it delivered?
Hosted is the flagship runtime: an isolated Meridians VM that runs the Program continuously. The focused mobile path and Stories handle return visits and consequential judgment; the desktop Studio opens the same expert in depth. License lets you self-host the identical daemon and canonical record. Email or messaging may become delivery adapters later, but they are not the product spine.
Does it work with my existing pipeline?
Yes. Bring source material in as text or research, connect supported sources, and use MCP or the operation surface to read and steer the same model from other tools. Export remains a first-class right. Integrations call Meridians' typed contracts; they do not write around the reducer or create a second state store.