Plans

Pensieve plans are based on your Context's footprint: the customer inventory you choose to connect. That gives you a direct dial over the thing you pay for. It does not turn model calls, tokens or OpenRouter prices into a second bill.

Anyone can create a context. A new context starts on No plan, so it does no work until a trial or a paid plan starts. Billing opens once it is created, with two ways to start:

• Book a call from the Trial card. We set Pensieve up with you, then start a 30-day free trial on that context, with no card required. The trial gives full access, with an allowance we size with you on the call.
• Choose a plan and pay upfront.

Creating a context never starts a trial, and each context can have one. If you named a website when creating the context, Pensieve connects it once the trial or a paid plan starts. There is no permanent free tier. When a trial ends without a plan, connector syncing, MCP access and AI work pause until you choose one. Your Pages and stored data are kept and stay readable in the app.

One flat rate: 20p per unit per month, in £100 steps from 1,000 units (£200) up to 10,000 units (£2,000). Plans differ by the support that comes with them. Above 10,000 units, pricing is agreed with you. Managed inference is included.

The slider only stops on positions the backend can actually sell, so the figure beside the thumb is the price that will be charged.

What counts as a knowledge unit

Pensieve measures the current knowledge it maintains. One knowledge unit is about 10,000 characters of maintained knowledge. Each source is weighted by the kind of knowledge it holds, and Pensieve adds up the weighted amounts across the whole Context before rounding once, so splitting the same content into more files does not increase the footprint.

These are billing representation weights: Literal, Tabular, Distilled, and Source Tree. They do not vary by vendor, so Notion and Google Drive do not get different prices for equivalent knowledge. Generated Pages, embeddings, revisions and processing structure do not add a second footprint.

Saving a connector selection is not gated on knowing what it will cost. Pensieve stores the selection, starts the sync, and measures each source exactly as ingestion renders it. Capacity is checked at admission, against the paid allowance or, during a trial, the trial's size.

Settings → Billing shows the exact used and allowed footprint for the Context, and Pensieve notifies the context's members as it passes 50%, 80%, 95% and 100% of its allowance, so the ceiling is never a surprise.

Untouched content is discounted

Content its source reports untouched counts for less: full weight for the first thirty days after its last edit, 75% from thirty days, and half from ninety. The clock is the source's own last-edited date, so an archive counts at the reduced weight from the day it is connected, and editing a page counts it fully again. Pensieve only reduces where the source gives an honest edit date — Notion, Google Drive, Linear and HubSpot do, and meeting transcripts cannot change once captured. Uploaded files, website pages and sources with no reliable edit date stay at full weight however long they have looked the same. Nothing is moved or hidden to earn the rate: the content stays fully live in Pages and search.

What happens at the allowance

On a trial or paid plan, reaching the knowledge-unit allowance pauses future connector runs, including updates to sources you have already connected. Knowledge already in the Context stays readable and searchable, Pensieve keeps curating it, and MCP access remains available. Connecting or restoring knowledge requires enough allowance. Remove connected knowledge or move to a larger plan to make room: syncing resumes automatically, and any background ingestion that was waiting is released.

After a trial or paid service period ends, incoming-data processing, connector syncing, MCP access and AI work (including chat and automatic curation) pause until payment restores access. A trial also carries an AI budget, separate from its knowledge units. If that runs out first, syncing and AI work pause while MCP access continues until the trial ends; choosing a plan, or asking us to raise the budget, resumes them. The app, manual Page editing, search, local Page sync and Git Mirror remain usable. An administrator pause stops syncing while MCP and funded chat remain available during the current trial or paid period.

There is no metered overage, inference-credit top-up or surprise usage charge. The allowance is a hard, visible boundary.

Usage and provider costs

Settings → Usage shows token activity so you can understand how the Context is working. It is operational information, not a cash forecast or balance.

Pensieve owns and operates the OpenRouter credentials used for inference. Provider receipts and model costs are reconciled internally as Pensieve's cost of serving the plan; they do not reduce your footprint allowance and are not passed through request by request.

Billing periods

Plan service windows run month to month from the Context's billing anniversary. For a paid self-service subscription, Stripe proves that each interval was paid and Pensieve keeps the exact plan, allowance, price and currency bought for that subscription. A later public price change does not silently reinterpret an existing purchase.