Frequently asked questions

/ˈtoʊ.pə.lɒɡ/ · TOH-puh-log.

£9.99/seat/month or £59.94/seat/year (equivalent to £4.99/seat/month: pay six, get twelve). Same per-seat rate solo or team. Each seat includes 1,000 credits at signup, pooled across the team and never expiring.

The subscription unlocks the product (canvas, scheduler, execution, all views). Credits buy AI calls. They're billed separately so you only pay for the AI you actually use.

No. The 1,000 credits per seat are a one-time grant when the subscription starts. They don't refill monthly and they don't refill on annual renewal. They also never expire, so the balance stays with you as long as your account exists. When you push harder, top-up packs (from £2.49) add credits that pool with the rest and likewise never expire.

Most users won't. The 1,000-credit per-seat grant comfortably covers many months of typical use. If you do run low, top-up packs start at £2.49 and never expire.

A 7-day free trial, 250 credits, no card required. Enough for around 5 plans (Express or Structured both cost 50 credits flat) plus a few decompositions and discussions.

Yes. Monthly and annual subscriptions can be cancelled in the customer portal. You keep access until the end of your billing period; credits stay in your account afterwards.

No. The scheduler is pure deterministic TypeScript. Topological sort, bin-packing, deadline propagation, critical-path analysis: all algorithmic. AI only writes the one-line plan summary, never the schedule itself.

It carries forward to the next day's queue, ranked first. No streak break, no shame. The schedule silently rebalances.

Yes. When a task is bigger than a day's available block, the scheduler splits it. Both halves keep the same task id and visually link with arrows in the day/week views.

A task that depends on a third party: your broker, your bank, a supplier. External tasks are visible in your dependency graph and gate downstream work, but they're not scheduled into your hours. Mark them done when the third party delivers.

Always. AI proposes structure; you sign off before anything commits. Decompose actions wait for your accept/reject. Inline editing on every node and edge.

Mistral. We use two tiers in different parts of the pipeline depending on the cognitive shape of the call: Mistral Small for constrained Q&A (yes/no, pick-one, score) and Mistral Medium 3.5 with adjustable reasoning effort for generative decomposition (break a goal into milestones). Medium 3.5 matched what we used to use Large for at lower cost, so Large is no longer in the pipeline. We call the Mistral API directly; no third-party AI middleware sits between you and the model.

Your dependency graphs, schedules, and observations stay in your Supabase row, scoped by RLS. Mistral receives only what's needed for the immediate request and is never used to train any model.

Topolog is web-first. A dependency graph is a two-dimensional artefact: panning a canvas with one thumb is genuinely worse than panning it with a cursor, and the IDE has nine analytical panels that need real screen real estate. The web app is responsive so you can read your plan and mark tasks done on a phone, but the authoring surface is built for the desktop.

Three panes: an explorer on the left (your goal, milestones, sentinels, floating facts), a canvas / source / analytics pane in the middle (nine tabs: Graph, Source, Parameters, Causal Threads, Spectrum, Critical Path, Parallelism, Pareto, Money), and an inspector on the right where the AI proposes the next move and you sign off. A bottom drawer surfaces validator problems and the Money panel. Everything is one keystroke away.

Yes. The canvas, scheduler, and execution views all work on any dependency graph, whether AI-generated or hand-built. On the graph you can click empty canvas to add a node, drag from one node onto another to connect them, click a node or edge to edit it in the inspector pane, and press Delete to remove the selection; or drop into the Source tab and author the whole plan in TOL directly. The deterministic scheduler runs whether the graph came from Express, Structured, or your keyboard.