AI memory comparison
Tephra vs Pieces
Pieces continuously captures your working context - code you copy, screens you view, audio you hear - and keeps it on your own device, reachable by MCP clients on that machine. Tephra saves only what an AI deliberately chooses to save, and serves it to any connected app from anywhere. Neither has a free tier. The real choice is between automatic capture on one machine and deliberate memory everywhere.
Sources read and prices checked 2026-10-10.
Obsidian Metrics makes Tephra and Tracker, the two products this page compares against others. This page is written by the maker of one of the products compared, so read the competitor columns against their own pages, linked below each table. Competitor facts come only from the vendor pages listed under Sources, read on the date shown above; vendors change plans and prices without notice. We are not affiliated with any company named here, and all product names belong to their owners.
Tephra and Pieces side by side
| Feature | Tephra | Pieces |
|---|---|---|
| What it is | A hosted memory your connected AI apps share: they recall from it and write to it over MCP or REST. | "The memory layer for modern work." A desktop app plus a local service whose Long-Term Memory engine continuously captures context from your workflow - code you copy, screens you view, audio you hear - and stores it on your own device. [1] [2] |
| Entry price | Tephra needs a paid plan: Premium at $19 a month in the Obsidian Metrics Skool community or $29 a month direct, or a $399 one-time seat. No free tier. | No free tier. Its own pricing FAQ asks "Is Pieces free to use?" and answers "No." Pro is $18.99 per user a month and Enterprise $22.99 per user a month, both billed monthly, with a 7-day trial and a 14-day money-back guarantee. Read 2026-10-10. [1] |
| How usage is limited | No usage metering. There are only abuse limits: 60 requests a minute per key and 120 a minute per member. | Per seat and otherwise unmetered: Pro lists "Unlimited MCP" and lets you switch between models under one plan. No token or request metering is documented. [1] |
| Reach across AI apps | One memory for every AI you connect, led by OAuth sign-in: the Connect page has steps for Claude (web, Desktop, mobile), ChatGPT, VS Code Copilot, Cursor, Claude Code, Gemini CLI, Windsurf, Zed, Mistral Le Chat and Perplexity, plus generic steps for any other MCP app, and a personal key as a fallback, plus REST. AI tools that support neither MCP nor REST calls cannot connect. Tephra's OAuth discovery endpoints were verified live on 2026-10-10. | Machine-bound: the MCP server is PiecesOS running on your own computer, which its docs call a required dependency. It names Claude Desktop, Claude Code, Claude Cowork, Cursor, Codex CLI, GitHub Copilot, VS Code, Zed, Windsurf, JetBrains, Raycast and more. There is no hosted endpoint and no OAuth; its docs do describe an ngrok setup to "share PiecesOS for remote access", so remote use means tunnelling your own machine rather than a vendor-hosted server. [3] [2] |
| Seeing and correcting what is stored | A member can view memories, edit or delete them, read edit history, contest a memory, export their own data (app session), and revoke any AI key with an activity log. | Good on control, less clear on editing. There is a browsable timeline of captured activity, you can pause capture for a set period, disable long-term memory entirely, exclude individual apps as sources, and clear all stored memory in one action. Its docs note that excluding a source stops new capture but retains what was already captured. We did not confirm per-memory editing or version history. [2] [1] |
| Hosting and self-hosting | Hosted service run by Obsidian Metrics. Self-hosting is not part of the product. We make no encryption-at-rest or compliance-certification claim. | Closed source but local-first, which is an unusual pairing. Its docs state that the on-device engine "runs entirely on your device" and that "All captured data is processed and stored locally. Nothing leaves your machine unless you explicitly choose to share it." We found no compliance certification on the pages we read. There is no self-hosted server to run and no licence to inspect. [2] [1] |
| How memories get in | The connected AI writes them by calling a remember tool. Tephra does not capture chats automatically and has no Gmail, Slack or Drive connectors. You can upload text, markdown, JSON, XML and text-layer PDF files. | Fully automatic and continuous, with no tool call at all: its engine captures code you copy, screens you view and audio you hear, and its sources include browser history, email, notes and meetings. It also has a Google Calendar connector. [2] |
Where Pieces is better
- Pieces. Captures without asking anything of the model or the user. Tephra holds only what an AI decided to save; Pieces holds the screens you looked at, the code you copied and the meeting you sat in, so for "why did we decide this three weeks ago" it has an answer and Tephra may have nothing. It keeps data on your own machine by architecture rather than policy - its docs say nothing leaves your machine unless you choose to share it - while Tephra's encryption at rest is not confirmed. Its consent controls - a pause for a duration you choose with a "Paused until" label, auto-stop on screen lock, per-app access control, and a clear-data dialog you can filter by time period, modality and app source - are more concrete than an inspect-only view, and it ingests screen, audio, email and meetings, which Tephra cannot. [2] [1]
Choose Pieces if: You want memory you never have to think about, you want it to stay on your own hardware, and you work mainly on one computer. Pieces captures what you actually did, where Tephra only holds what an AI decided to write down.
Where Tephra is better
- Connected AI apps read and write the same memory directly, rather than moving it between apps by pasting text. Claude's import and export, for example, works by pasted text, and its import path is marked experimental.
- A flat price with no usage metering, instead of credits or request quotas.
- A shared library that Tephra grows every hour from public research and regulatory sources, included with the product. We did not find a competitor that bundles one; that is a qualitative check, not an exhaustive market survey.
- Recall in the app tells you which related memories it did not load and how complete the answer looks.
Choose Tephra if: You need memory reachable from more than one machine - a phone, a web client, a server-side agent or a team - because Pieces' MCP server is PiecesOS running on your own machine, which its docs call a required dependency, with no hosted endpoint and no OAuth. Its only documented remote path is tunnelling your own machine with ngrok.
Where Tephra falls short
- Tephra has no free tier and no self-hosted option.
- The MCP tools return ranked records only. The completeness alert is shown in the app, not through MCP.
- No automatic chat capture, and no connectors to Gmail, Slack or Drive.
- No image or audio ingest, and no OCR for scanned PDFs.
- Stale-fact handling is partial.
- No published, vendor-comparable benchmark, and no accuracy lead we can defend.
What we have and have not measured
On 50 held-out LongMemEval-S questions never used in Tephra development, Tephra answered 39 of 50 correctly (78%, 95% confidence interval 65-87%) with a gpt-4o-mini reader, and every evidence session was in its top 10 results for 46 of 46 answerable questions.
Method: LongMemEval-S cleaned, stratified 100-question subset; gpt-4o-mini reader over top-10 retrieved items; official GPT-4o judge prompt; one run, 2026-10-08; held-out questions are the 50 never used in earlier Tephra tuning; competitors run locally from open source under the same harness.
In that same-harness run, none of Tephra's leads over the other systems tested was statistically significant after correcting for multiple comparisons, and its margin over a plain keyword-search baseline was not significant either. We do not claim Tephra is the most accurate memory. We did not benchmark Supermemory, Zep, ChatGPT memory or Claude memory, and we do not compare ourselves with any vendor's self-reported numbers.
Frequently asked
Is Pieces free?
No, and its own pricing FAQ says so: it asks "Is Pieces free to use?" and answers "No." Pro is $18.99 per user a month and Enterprise $22.99, with a 7-day trial and a 14-day money-back guarantee, as read on 2026-10-10. Tephra has no free tier either, so this is not a difference between us.
Does Tephra record my work automatically like Pieces does?
No. Tephra saves only when a connected AI calls its save tool, or when you add or upload something yourself. The server instructs connected AIs to save durable facts and decisions, but it cannot force them to, and nothing is captured from your screen, your audio or your email. Pieces captures continuously; if that is what you want, it is the better tool.
Where is each one's data kept?
Pieces states that its engine "runs entirely on your device" and that "All captured data is processed and stored locally. Nothing leaves your machine unless you explicitly choose to share it." We found no compliance certification on the Pieces pages we read. Tephra is cloud-hosted on Cloudflare; your memories are tied to your account and are not visible to other members, and we do not claim encryption at rest because we have not confirmed it.
Did you benchmark Tephra against Pieces?
No. We have not benchmarked Pieces and we make no accuracy claim against it.
More comparisons
Sources
Competitor facts come only from the pages below, accessed 2026-10-10.
- Pieces pricing: https://pieces.app/pricing (accessed 2026-10-10)
- Pieces documentation, long-term memory: https://docs.pieces.app/products/core-dependencies/pieces-os/long-term-memory (accessed 2026-10-10)
- Pieces documentation, MCP: https://docs.pieces.app/products/mcp (accessed 2026-10-10)