Pinning Context

Every chat conversation has a context the agent can see — your active document, your recent editor activity, the most recent conversation history. This default context is usually enough. But sometimes you need the agent to be paying attention to something specific that isn’t naturally in the default context, and you don’t want to keep explaining it in every message. That’s what pinning is for. A pinned item is attached to the conversation as context that the agent sees on every message, regardless of what else is happening. Pin a lore entry about a character and every message in the conversation will have that character’s details fresh in the agent’s mind. Pin a research bookmark and every message will be grounded in that research. Pin a document and the agent will reference it naturally. The pin system is how you shape the agent’s context without having to re-explain yourself.

This matters because LLM context is finite and the agent’s memory of the conversation is only as good as what’s actually sent in each request. If you mentioned a character three turns ago and the conversation has moved on, the agent might have lost that detail to context compression. Pinning keeps the important stuff stable.

What can be pinned

Several types of content:

Lore entries

Any entry in your Legendry can be pinned. Once pinned, the entry’s content (title, summary, sections, tags) is included in the chat’s context on every message. The agent can reference the entry by name, quote from its sections, or apply its canon to the conversation without you reminding it.

Typical use: pinning a character entry when the whole conversation is about that character. “Let’s talk about Kent’s arc in book two” works better if Kent’s lore entry is pinned so the agent has his full history accessible throughout.

Documents

Manuscript documents can be pinned — when pinned, the document’s content (or a relevant excerpt) is available to the agent. Useful when you’re workshopping a specific scene and want the agent’s responses grounded in the actual prose you wrote.

Limit: pinning a very long document can exceed context windows. For full manuscripts, pin specific scenes or chapters rather than the whole document. The agent can pull more on demand if it needs to.

Research bookmarks

Smart bookmarks from your research library can be pinned. Useful when you’re discussing a topic you’ve already researched and want the bookmark’s content available without re-running a search on every message.

Typical use: pinning a bookmark about medieval siege tactics when you’re writing a siege scene and want the agent to have those details readily available.

Plotlines

Plotlines from Plot Studio can be pinned. The agent sees the plotline’s structure, beats, plants, and promises. Useful when you’re debating a plot decision and want the agent to see the full arc, not just the current scene.

Stat blocks

For mechanics-heavy discussions, stat blocks from your Magic System can be pinned to make them available for the duration of the conversation.

How to pin

From the chat interface, click the pin icon next to the message input. A picker opens showing your available content organized by type:

Lore entries — searchable list of all entries in your Legendry.
Documents — list of your writing documents.
Research — list of smart bookmarks.
Plotlines — list of active plotlines.
Stat blocks — list of stat blocks attached to lore entries.

Pick one or more items. They’re added to the pinned list for the current conversation. Each pinned item appears as a chip above the message input, showing its title and a remove button.

You can pin up to a configurable limit per conversation (default 5). More than 5 starts to eat into the context window aggressively, so there’s a soft limit that warns you before adding more. Hard limit is 10 — past that, the chat will refuse to add more pins until you remove some.

The visual display

Pinned items are visible in two places:

Above the message input. A row of chips showing each pin, with a remove button. This is the “currently pinned” indicator — you can see at a glance what the agent is seeing.
Per-message in the chat history. Each message shows which pins were active when it was sent, so you can tell whether a particular response was informed by specific context.

If you remove a pin mid-conversation, it’s gone from subsequent messages but the history of previous messages still shows that the pin was active when they were sent. This matters for retracing why a specific response was what it was.

Per-conversation scope

Pins are per-conversation, not global. Every chat conversation has its own pin list, and starting a new conversation clears the pins.

This is deliberate. Different conversations are about different things, and carrying pins from an unrelated conversation would bloat the context. If you always want the same pin in every conversation, the right answer is usually to make that content part of the agent’s default context via the Skills system, not via manual pinning.

Auto-pinning

Some context gets pinned automatically without user action:

Active document. When you open a chat from within the editor, the currently-open document is auto-pinned. Close the document or start a new chat and the pin clears.
Selected lore entry. When you open a chat from the Legendry detail panel, the selected entry is auto-pinned.
Current research page. When you chat while the Research browser has a page loaded, the page content is auto-pinned.

Auto-pinned items are visually distinguished from manual pins — usually a slightly different color chip or a lock icon. You can remove an auto-pin manually if you don’t want it in the conversation, but the default is that they’re there because the context of how you opened the chat suggests they should be.

The relationship to Skills

Skills are the other way to add persistent context to a chat — but they work differently. Skills are global, configured once, and apply to every conversation with the agents they target. Pinning is per-conversation, temporary, and requires explicit action.

Use skills for content that applies to every conversation: style guides, genre conventions, character pronoun rules. Use pins for content that applies to this specific conversation: the character you’re debating, the scene you’re workshopping, the research you’re integrating.

Skills and pins layer together — an agent with three active skills and three pins has all six context sources available simultaneously.

Context window impact

Pinning isn’t free. Each pin adds to the context window on every message, which can hit limits in long conversations or with verbose content. The chat panel shows a “context usage” indicator that fills up as your conversation and pins consume tokens:

Green: plenty of room, no compression happening.
Yellow: approaching the window limit, some compression starting.
Orange: heavy compression, older messages getting trimmed.
Red: near the limit, aggressive compression. Remove pins or start a new conversation.

If you see orange or red, the chat is working but you’re losing some of the earlier conversation to compression. For long-running conversations with heavy pinning, consider splitting into multiple chats instead of fighting the compression.

What pinning isn’t

Not a way to bypass context windows. If you pin a 50,000-token document in a conversation with a 32k-token model, the pin will be trimmed or cause errors. Pins respect the window limit — they’re just prioritized inside it.
Not a training mechanism. Pinning doesn’t teach the agent anything persistent. It just makes content available during the current conversation. Closing the conversation clears the pin’s effect.
Not a substitute for good entity detection. If your prose correctly mentions a character by name, the agent often picks up on the mention without needing the character pinned. Pinning is for when the default context isn’t enough, not for every character reference.

What’s next

Chat overview The chat interface where pins are managed.

Skills The other persistent context mechanism for agent conversations.

Writers' Room How pinning interacts with multi-agent mode.