Just want a personal AI knowledge base? See the personal-use comparison →
TroveFiles vs. Obsidian: one is for humans, the other is for agents.
Obsidian made it cool to organize your knowledge in a folder of markdown files. Now agents are doing the same thing in the cloud. Here's how the two tools differ, and how they fit together.
Different audiences,
same filesystem.
Obsidian is built for a human to open an app and browse their notes. TroveFiles is built for an agent to read and write files programmatically. The user is the one calling the shots: a human in Obsidian, an agent in TroveFiles.
Humans can also open the TroveFiles dashboard in any browser to see what the agent wrote (markdown rendered, images and PDFs previewed, audio and video playable, with snapshots one click away). But the heavy lifting of reading and writing happens through the agent calling the API. Both tools sit on the same primitive (a folder of files); the difference is who's holding the keyboard.
One opens an app. The other speaks an API.
Obsidian
A markdown viewer with a plugin system. It renders .md files from a folder on your local machine into a browsable, linkable interface. Graph view, backlinks, templates, community plugins. All designed to help a human navigate and connect their notes.
When people say "Obsidian + Claude Code," what's actually happening is Claude Code reading and writing files on the local filesystem. Obsidian just renders them nicely afterward. Obsidian is the frontend. The filesystem is doing the real work.
TroveFiles
A cloud-managed POSIX filesystem for AI agents, with a web dashboard humans can open in any browser to see what the agent wrote. The agent gets three tools: trove_exec, trove_read, trove_write, plus the full Unix toolchain preinstalled (grep, awk, sed, pdftotext, ffmpeg, jq, exiftool).
The human side is a real file browser at trovefiles.dev/dashboard: sortable file tables, markdown rendered with GFM, images and PDFs previewed inline, audio and video playable in place, snapshots with one-click rollback. No installation, works on any device with a browser.
Each customer gets an isolated namespace. Files persist across sessions. Daily auto-backups (last 7 kept) plus manual save points you can roll back to in seconds. HMAC-signed webhooks fire on every write. No outbound network from the sandbox.
Three names for
the same pattern.
Three names for the same pattern have emerged across the ecosystem. The LLM needs a context file to read before it does anything else.
CLAUDE.mdis Claude Code's project-context file, read at the start of every Claude Code session.AGENTS.mdis the vendor-neutral version, natively supported by Cursor, GitHub Copilot, OpenAI Codex, Windsurf, and 20+ other tools. Stewarded by the Agentic AI Foundation under the Linux Foundation as of December 2025.- A per-session handoff file is the TroveFiles equivalent. The agent writes it at session end, the next agent surfaces it on first turn, and the protocol enforces the loop without per-app prompt engineering.
Same pattern, three forms. Where the file lives (local filesystem, agent-tool root, or cloud workspace) determines whether the system works for one person on one laptop or for thousands of users across a platform.
The honest comparison.
| Dimension | TroveFiles | Obsidian |
|---|---|---|
| Built for | Agents and the humans who manage what they wrote | Humans browsing notes |
| Primary user | The agent (humans use the dashboard to inspect and roll back) | The human at the keyboard |
| Storage location | Cloud, per-tenant namespace | Local filesystem |
| Access | Web dashboard for humans + API and shell for agents (via MCP / SDK) | Desktop app (GUI) |
| Snapshots / time travel | Built in. Daily auto-backups (last 7 kept) + manual save points + one-click rollback | 1 year of version history with Obsidian Sync; otherwise via git or third-party plugins |
| Multi-tenant | Yes, isolated namespaces with scoped API keys per tenant | No native multi-tenancy (Obsidian Sync shared vaults allow team collab on a single vault, not many isolated tenants) |
| Unix toolchain | Full toolchain preinstalled (grep, awk, sed, pdftotext, ffmpeg, jq, exiftool) | None |
| Webhooks | Yes, HMAC-signed on every write | No |
| Cross-device sync | Cloud-native by default (any browser, any device, any AI client, any session) | Local + Obsidian Sync (syncs across one human's devices or shared vault members) |
| Security model | Sandboxed, no outbound network, process isolation | Local filesystem trust |
| Plugin ecosystem | None (the shell is the plugin system) | Thousands of community plugins |
| Graph view | No | Yes (the headline Obsidian feature) |
| Best for | Cloud-native agent + human collaboration at scale | Local-first personal knowledge management |
Pick TroveFiles when…
- • You want cloud access from any browser, on any device, with no app to install.
- • You want time travel. Daily auto-backups plus manual save points, with one-click rollback if the agent breaks something.
- • You're an individual using Claude Desktop or Cursor and want persistent memory across sessions and devices.
- • You're building a product for users. Each customer needs their own isolated, persistent workspace with scoped API keys.
- • Files need to trigger workflows. HMAC-signed webhooks fire on every write.
- • You want the agent to read non-markdown files (PDFs, audio, video) without installing local CLI tools yourself. Real Unix tools (pdftotext, ffmpeg, exiftool) are preinstalled in the agent's sandbox.
Pick Obsidian when…
- • You're the only user. Personal wiki, daily notes, one human, one vault.
- • You want to browse visually. Graph view, backlinks, canvas. Powerful tools for a human making connections.
- • The agent is a local assistant on your laptop. Claude Code on the same machine as the vault works fine.
- • You want a deep plugin ecosystem. Thousands of community plugins for every workflow.
Some personal setups use both. TroveFiles as the persistent agent workspace. Obsidian as a local viewer for graph view and canvas. The agent writes to TroveFiles, you sync those files to a local Obsidian vault. See the memory use case for how this looks in production.
Obsidian vs. TroveFiles,
answered.
What is TroveFiles, architecturally? Is it just MCP-over-S3?
No. Three layers stacked. (1) Storage: per-namespace cloud storage that presents to the agent as a normal POSIX filesystem. (2) Execution: every command runs as /bin/sh -c <command> inside a sandboxed Linux container with kernel-level filesystem confinement (the agent can only touch its own namespace), network egress blocked, and a separate UID for the exec processes. About 30 CLI tools preinstalled (pdftotext, ffmpeg, jq, awk, grep, exiftool, python3, imagemagick, the GNU coreutils, etc.). (3) Protocol: REST API plus an MCP server with built-in agent-memory discipline. The shell is the differentiator: the agent gets bash, not a custom DSL or a query API. Whatever a Unix person can do with grep / awk / ffmpeg at a terminal, the agent can do too.
How does the cross-session memory loop close itself?
The TroveFiles protocol is opinionated about agent memory. It guides the AI client to surface a session-start orientation block (recent files, workspace summary, prior handoff) on first turn, and to write a fresh handoff note before ending substantive sessions. This is the difference between TroveFiles and a generic MCP-over-storage server: the memory discipline is baked in, so the loop closes itself in production without app-side prompt engineering on every turn.
I love Obsidian. Is TroveFiles trying to replace it?
No. They're built for different users. Obsidian is for a human opening an app and browsing their notes. TroveFiles is for an agent reading and writing files through an AI client. Most people who use both keep using both. The agent works in TroveFiles; the human reads in Obsidian.
I'm running Claude Code with my Obsidian vault today. Should I switch?
Not unless you need files accessible beyond the laptop the vault lives on. Local Obsidian + Claude Code is a great single-machine setup. The moment you want the same agent and the same files reachable from a phone, a web app, a CLI, a scheduled job, or any client other than the laptop you started on, you need cloud-native infrastructure. That's when TroveFiles fits.
Can I use both at the same time?
Yes. The agent writes to TroveFiles. You sync those files to a local Obsidian vault and browse them with graph view, backlinks, and canvas. Agent writes, human reads.
Does TroveFiles have a desktop app like Obsidian?
Not a desktop app, a web dashboard. Open trovefiles.dev/dashboard in any browser and you get a real file browser: sortable file tables, markdown rendered with GitHub Flavored Markdown, images and PDFs previewed inline, audio and video playable in place, snapshots panel with one-click rollback. No installation, works on phone or laptop or tablet. It's built for managing what the agent wrote (and rolling back when it writes something wrong) rather than the long-form note-taking Obsidian is designed for, but the file-browsing experience itself is fully featured.
How is this different from Obsidian Sync?
Obsidian Sync syncs your vault end-to-end encrypted across your own devices, gives you a year of version history per file, and supports shared vaults. What it doesn't do is solve cloud-native, AI-reachable storage at product scale. There's no hosted remote API. Agents only reach the files when they're sitting on a device where Obsidian, Claude Code, or a community MCP server is installed and running. There's no per-tenant isolation, no scoped API keys, no webhooks, no shell sandbox for the agent to run real Unix tools in. Obsidian Sync handles cross-device sync for one human or a small team. TroveFiles is the cloud-native, AI-native, multi-tenant filesystem you build a product on top of.
What about Obsidian's plugin ecosystem?
Keep it. Obsidian plugins are designed for the human-browsing experience (graph view enhancements, daily notes templates, dataview queries). TroveFiles operates on the agent side, where the equivalent of a "plugin" is just another shell command the agent already knows. The two ecosystems don't compete; they sit on opposite ends of the same filesystem.
If I'm building an agent product, do I still need Obsidian in the stack?
No. Obsidian is a personal productivity tool your end users might happen to use on their own machines. It's not infrastructure for a product. If you're building agents that serve thousands of customers, each customer needs their own isolated workspace, scoped API keys, audit trails, and webhooks. That's TroveFiles, not Obsidian.
Who's running TroveFiles in production?
TroveFiles is the storage layer behind Silvia, our AI CFO with over $30 billion in connected assets. Every Silvia user has a TroveFiles namespace where the agent stores memories, skills, and preferences and retrieves them via shell commands across sessions.
Give your agent a real
filesystem to work in.
Free tier, no credit card. Three lines of Python, or one MCP install command into Claude Desktop.