Petals vs Claude memory

Claude is genuinely excellent

Say it plainly: Claude's writing is the best of any major model. Coding, research, long-document analysis — the benchmarks confirm what people who use it daily already know. Claude Opus 4.6 scores 80.8% on SWE-bench Verified; 91.3% on GPQA Diamond. Developers building serious things routinely reach for Claude first. The model quality is not in question here.

This page is about memory, specifically — and one other thing: whether your assistant can act on what it knows.

How Claude remembers

Claude runs three memory layers.

Chat Memory (global) synthesizes your standalone conversation history into a wiki-style document organized into sections: Role, Work Context, Communication Preferences, Technical Preferences, and similar. The synthesis runs roughly every 24 hours, though you can trigger an immediate update by telling Claude to remember something mid-conversation.

This is a real structural advantage over ChatGPT. Claude's wiki is readable top-to-bottom in a single pass — you can audit it in a few minutes. You can open Settings → Capabilities → Memory, read every section, and edit anything that's wrong. Claude also tells you when it is drawing on a memory at conversation start, which is a design choice worth noting. Compared to ChatGPT's Dreaming V3 — which synthesizes inferences from years of history that its own summary page admits may not show everything it remembers — Claude's wiki is more auditable. That is a genuine transparency win, and it is fair to say so.

Chat Memory updates via standalone conversations only. Project chats feed a separate, isolated project memory and do not contribute to the global pool.

Projects (Pro and above) give each project its own memory space with uploaded knowledge files, a 200K token context window, and RAG fallback when the project knowledge exceeds that window. File formats include PDF, DOCX, CSV, and about a dozen others. Projects are paywalled at $20/month Pro — free users get Chat Memory but not Projects.

Incognito Chats opt out entirely: nothing stored, nothing synthesized.

One honest limitation: the 24-hour synthesis cycle means something you say today may not appear in memory until tomorrow. And global Chat Memory and Project memory are completely separate — there is no unified view across both. If you move between project and standalone conversations, context fragments.

How Petals remembers

Petals builds an entity-relation knowledge graph. When you tell Petals something, it extracts nodes (people, places, events, ideas), relationships between them, and individual claims — each claim linked to its source conversation. You can browse the graph, click into any node, see exactly which statements came from which conversations, and edit or delete anything.

This is structurally different from a wiki. A wiki tells you what Claude believes about you. A graph shows you the shape of what it knows — and lets you query it.

Contradiction detection is part of the design: when new information conflicts with an existing claim, Petals surfaces the conflict rather than silently overwriting it. Claude's wiki does not do this.

Memory ingestion today works two ways. Manual: paste notes, drop a file, link a document, submit a transcript — anything you can put into a text form can become part of what Petals knows. Automated: the ingestion API accepts documents and transcripts asynchronously via POST requests and processes them as background jobs. Connecting an always-on data feed requires writing a small script against that endpoint and authenticating with an API key from Settings → API keys. This is a pattern for the technically comfortable, not a turn-key integration.

Where Claude is better

Model quality and writing. There is no version of this comparison where Petals claims its underlying models are better than Claude Opus. They are not. Claude is the model Petals users reach for when they want nuanced prose, careful reasoning, or serious code — and they can, because Petals lets you choose which model powers each assistant.

Projects with persistent document context. The 200K token project context window, file uploads, and RAG fallback make Claude Projects genuinely powerful for document-heavy workflows. If your use case is a research project with a stack of papers, Claude Projects is well-designed for it.

Structured memory import. Claude accepts memory exports from ChatGPT, Gemini, and Grok — a practical convenience if you are migrating. Petals does not currently offer a one-click import from competitors.

Native ecosystem. Claude is built by Anthropic, ships as a first-party app, and has a polished mobile experience. For users who want a single, maintained product from a single company, that simplicity has real value.

Memory acknowledgment at conversation start. Claude explicitly tells you when it is using a memory. This is a small design detail that builds trust in a way most assistants skip.

Where Petals is different

A graph instead of prose. Claude's wiki is a document. It tells you what the model believes about you in sentence form. Petals' graph is a structure — nodes, edges, claims, sources. You can navigate it. You can ask "what does Petals know about Ana?" and get a set of connected facts with their origins, not a paragraph.

Assistants that act. Claude remembers. It does not take actions on your behalf. Petals has connectors and automations: your assistants can reach into tools (calendar, email, Notion, and others via Composio), propose scheduled tasks, and run them with your approval. Morning briefs, reminders, follow-ups — the memory is not just context for conversation, it is the foundation for action.

Price. Petals Plus is $9/month. Claude Pro is $20/month. Both include memory. The difference is what you get with it.

Model flexibility. Petals is not tied to a single provider. You can run Claude, GPT, Gemini, or your own API keys inside Petals. The memory graph is the constant; the model behind any given assistant is your choice.

The honest summary

If you want the best writing model and a clean, well-audited memory wiki — Claude is excellent and fairly priced at $20/month for what it does.

If you want to see the shape of what your assistant knows, navigate it like a map, have it notice when facts contradict, and have assistants that reach into your apps and act — that is what Petals is for.

If that sounds like what you need, start free and see what the graph looks like after your first few conversations. The pricing page has the full tier comparison.