Use AI without hitting
your usage limits
How long your Langdock allowance lasts comes down to two things: which model answers, and how much text it has to read and write to do it. More counts as text than you would think. Not just your message, but the whole chat so far, every attached file, every search result and the answer itself. It is counted in tokens, small pieces of text.
By Dennis Cutraro, Co-Founder and Managing Partner · As of:
How your allowance lasts longer
Use the smallest model that can do the job. Give it only what it really needs, and say exactly what should come out at the end.
1. Choosing the right Langdock model
Auto is the default, and for everyday work it is enough. Langdock then picks a model based on your first message. Choosing yourself pays off when you know the task is very small or very hard.
In Langdock you hover over a name in the model picker. A card shows what the model is good for and roughly how much it uses. You can click through the tiers here.
Model picker
The dots show usage. Go up a tier when the result is not good enough.
Tier
Auto
- Automatic selection
When does it make sense?
For everyday tasks. Langdock picks the model based on your first message.
Usage
Tier
Efficient
- GPT-6 Luna
- Claude Haiku 4.5
When does it make sense?
Short emails, translations, simple summaries and pulling information out of texts.
Usage
Tier
Balanced
- GPT-5.6 Terra
- Claude Sonnet 5
When does it make sense?
Writing texts and code, comparing information and working with several integrations.
Usage
Tier
Complex
- GPT-6 Sol
- Claude Opus 5.5
When does it make sense?
Analysing hard questions, weighing options, and solving complex data analysis and programming tasks.
Usage
Tier
Maximum
- GPT-6 Astra
- Claude Fable 5.1
When does it make sense?
Solving the most complex problems and planning, working through and checking long tasks with many linked steps.
Usage
Model examples as of September 2026. The scale is a guide, actual usage also depends on your task.
What each model costs per million tokens is listed in Langdock's model overview.
2. Saving tokens: phrasing tasks efficiently
- 01
Write short, clear and complete.
Skip the filler. Bundle the goal, the necessary information and the desired result in one message.
- 02
Set the answer length.
For example: "Five bullet points, 150 words maximum." For corrections, ask only for the changed sections.
- 03
Narrow down files and search.
Attach only the pages or chapters the task needs. For search tasks, say which folders or time ranges the AI should look through.
- 04
Start a new chat for a new topic.
Whatever you still need from the old chat, bring along as a short summary.
- 05
Combine templates with skills.
A skill is a saved workflow, for example an offer template plus a ready-made program that fills it. The AI then only drops in the new content instead of rebuilding the workflow every time.
- 06
Solve complex tasks with a strong model.
Sounds like the opposite of saving. But every correction round with a model that is too weak costs too, and in the end often more.
Example
"Summarise the offer: price, services, open questions. Five bullet points maximum."
3. Keeping an eye on the context window and usage
Next to the model picker in the input field sits a small circle. Clicking it opens a menu with two parts: the context window at the top, your usage in the five-hour session and in the week at the bottom, each with the next reset.
The context window is something like the model's desk. Everything in the chat sits on it, and for every answer the model reads the whole desk again. The fuller it is, the more each round costs.
A new chat clears the desk. It does not restart the session, though: your usage so far keeps counting.
In the replica here you can click every row. Drag a PDF into the chat and watch the desk fill up.
Message Langdock …
What this row means
System tools
Tools Langdock gives every chat, such as search and file access. Their descriptions take up room in the window even if you never use them in this chat.
You cannot change this. It does explain why even an empty chat does not start at zero.
System prompt
Langdock's fixed instructions to the model: tone, rules, format. Sits at the start of every chat, before your first message.
Messages
The history so far, your messages and all answers. Grows with every turn, because the model re-reads the whole history for every answer.
This is where you save the most: start a new chat for a new topic and bring the interim result along as a short summary.
Skills
The descriptions of all skills enabled in the workspace, so the model knows what it can call. So every enabled skill takes up room, even when you do not use it.
Files
Attached documents land in the window in full, every page, every table. A 40-page PDF costs more than the entire chat so far.
Attach only the pages or chapters the task needs.
More tools
A link enables the open URL tool. The tool description enters the window; the page content only once the model reads it.
Example values, simplified. The real numbers are in your chat.
Drop something into the chat
Grab, drag into the chat, release.
What the session limit is for
It puts on the brakes after five hours, so one long sitting on Monday morning does not eat the whole week's budget.
Tip: send the first message in the morning.
It starts your session. Write at 8 and the reset comes at 13, even if you take breaks in between. That gives you a fresh window in the afternoon. The weekly limit is not affected.
Langdock limit reached: what then
Then a more economical fallback model answers until the window resets. Unless your company has enabled extra usage.
If you need more, ask the person who manages Langdock at your company. Extra usage costs extra, so it is their call. How the limits work in detail is explained in Usage limits in Langdock.
Frequently asked questions
Sources
Author
Dennis Cutraro is Co-Founder and Managing Partner at Unfuture. He talks to customers about Langdock rollouts every day and is in ongoing contact with the Langdock team.
