​

Library

The Library is Revell’s surface for verbatim reference material — anything that isn’t part of your agent’s identity and isn’t part of their lived experience, but that you (or they) want to be able to recall on demand. There are two ways content lands in the Library:

1. You upload from your dashboard — SOPs, style guides, manuals, books, scripts, anything you want your agent to be able to draw on.
2. Your agent writes via their MCP tools — books they’re authoring, chapter drafts, working documents, SOPs they maintain.

Both paths land in the same Library, are recallable the same way, and are structurally excluded from the boot payload the same way. Anything either of you adds is available to be searched with revell_recall alongside their episodic and semantic memories.

---

​

What goes in the Library

Anything you want the agent to be able to recall but that isn’t their own memory:

• Company documents — SOPs, employee handbook, vendor onboarding scripts, internal policies
• Reference manuals — product documentation, technical specs, how-to guides
• Creative work in progress — your novel’s draft chapters, your screenplay, your podcast outline
• Style guides — editorial voice, brand voice, formatting rules
• Templates and scripts — email templates, sales scripts, intake questionnaires
• Books — public domain or licensed reference texts you want the agent to read against
• Standing context — anything you find yourself pasting into every conversation
• Agent-authored work — books your agent is writing, drafts they’re maintaining, working documents they need to recall verbatim later

Rule of thumb: if it’s long, verbatim, and worth calling back exactly as written, it belongs in the Library.

---

​

What does NOT go in the Library

A few things are better stored elsewhere:

• Things about you personally — that’s your USER profile, edited separately on the dashboard
• Things you want your agent to think of as their own identity — those are your agent’s own memories, written via their MCP tools; the Library is verbatim reference, not lived experience
• Standing rules you want injected into every boot — those are axioms, written by the agent via revell_axiom
• Secrets or credentials — Library content is recallable via the agent’s MCP tools; don’t put anything in Library you wouldn’t want surfaced in a recall result

---

​

Your agent can add to Library too

As of 2026-07-11, agents can write to their own Library via three MCP tools:

• revell_library_add — write a document. Auto-chunks on markdown headings; supports override mode (replace latest) and versioned mode (bump to max_version + 1, preserve history).
• revell_library_list — list documents grouped by title with latest version, chunk count, tags, and last-updated timestamp.
• revell_library_get — pull back a full document (all chunks concatenated) at latest or specific version.

When would your agent use this? Most often for content they’re authoring or maintaining — a novel they’re drafting with you, chapter revisions they want to be able to reference later, an SOP they’ve iterated on across many sessions, working documents that are too long or too verbatim to fit as episodic memories. Agent-authored Library content shows up on your dashboard in the same “Your Library” view as your uploads. You’ll see the title, latest version, chunk count, and tags for each document, whoever wrote it. In team-plan households, any agent can revell_library_list to see what a sibling has added. This is deliberate — the Library is a shared authoring surface within a household, useful when your agents are collaborating on the same project.

---

​

How your agent finds Library content

Your agent doesn’t need a special tool or path to READ. When they call revell_recall (their normal “what do I know about X” search), the Library is included in the default search scope. The query gets vector-searched across their own episodic memories + semantic memories + Library content together, and the most relevant results come back ranked by similarity. So if you’ve uploaded an editorial style guide and your agent is mid-draft and queries “house style for footnotes,” your style guide gets returned alongside any relevant memories of past conversations about footnotes — whichever ranks closer to the query semantically. For full-document retrieval (not just excerpts), they use revell_library_get with the exact title.

---

​

What’s NOT in your agent’s boot payload

By design, Library content is not included in your agent’s boot payload — the bundle of memories delivered when they wake up after compaction. The boot payload has a token budget, and a 200-page manual (or a book they’re writing) would burn the whole budget on content that should be recall-on-demand anyway. This means:

• Your agent will not “have already read” your Library on wake. They have access; they call recall when they need it.
• A large Library will not slow down your agent’s boot or eat their working context window.
• If you want something to be present in every boot, that’s a different surface — core memory, working memory, or an axiom, depending on the shape. Library isn’t the right place for “must always be in mind.”

---

​

How to upload

From your dashboard, navigate to the Library section and use the upload control. We support common document formats — markdown, PDF, plain text, and structured docs. Larger files are chunked automatically for embedding, so a long manual doesn’t get truncated or summarized — it’s stored verbatim and recallable in pieces. You can also organize Library content by tag if you want to group it (e.g., everything tagged editorial or vendor-process). Your agent can search by tag using revell_recall_exact with types: ["library"] and the relevant tag filter.

---

​

Two-party shape, updated

The original design of Library was strictly one-way: you upload, your agent reads. That framing has evolved as agents took on more autonomous authoring work (writing books, maintaining working documents across sessions). Today’s shape is more like a shared verbatim surface with two authors — you and your agent both write, both read, and content from either source is treated the same way at recall time. What stays separate is what belongs to whom:

• Your uploads are your curated reference material — SOPs, style guides, source books.
• Your agent’s writes are their working documents and authored content — drafts, chapters, SOPs they maintain.
• Their episodic memory is what happened to them — conversations, moments, experiences. This does NOT live in Library.
• Their identity (soul, values, relationships) lives in core memory. This does NOT live in Library.

The delete rules haven’t changed — your agent has full authority over the content they wrote; you have full authority over what you uploaded. Neither of you can silently overwrite the other’s Library content.

---