Built on Recalld memory · recall, not replay

Chat that remembers.

Every fact. Not every transcript.

Recalld Chat is a chat app built on the memory layer itself. Each turn recalls the few facts that matter instead of replaying your history, so conversations stay sharp after months of use and your context never silently overflows.

See the numbers
Free tier, no card · Runs on your Recalld credits · EU and US regions live

Memory you can see, scoped so it never leaks.

Most assistants either forget everything or remember everything into one big soup. Recalld Chat does neither.

  • scoped Each chat keeps its own memory; chats inside a project share that project's pool. What you say in one conversation stays out of unrelated ones by design.
  • rules Standing rules apply to every chat, and only change with your confirmation. The assistant proposes a rule, you click Add or Dismiss. Nothing is saved behind your back.
  • inspect Every reply shows what was recalled, from where, and what it cost, split into chat, tools, and memory credits. Open any message and audit it.
you › What did we decide about the Berlin launch?

chat › Moved to March 12, press event cut. Ana owns
the venue refund; the budget freeze ends Friday.

⌁ recalled 3 facts · scope: this project
⌁ this turn: chat 1,204 + memory 218 credits

you › From now on, always answer in bullets.

chat › Proposed as a standing rule. It applies
everywhere once you confirm.
[ Add rule ]  [ Dismiss ]

Everything a daily driver needs.

Projects and folders

Group chats into projects with a shared memory pool, or keep them standalone. Folders and soft delete keep the workspace tidy.

Your pick of models

Gemini and Claude models, switchable mid-conversation. History and memory carry over unchanged.

Web search built in

The assistant searches the web when a question needs it and shows its sources. What it finds can be remembered like anything else.

Costs in plain sight

A per-session stats bar and a per-message split of chat, tool, and memory credits. No other major chat shows you this.

Memory modes

Recall raw past turns with author and date, or distilled facts. Per-session setting with a per-user default.

Tone as a setting

Concise engineer, direct analyst, neutral, or creative. A plain preference, not something the model guesses at.

The economics

Your vector database isn't free.

Skip the managed memory layer and you still pay, three times: for the infrastructure, for the pipeline around it, and, biggest of all, for every fat context you stuff into your model.

the DIY stack

you run it

A vector database is the cheap part. The bill is everything you bolt on to make it behave like memory.

  • Vector DB hosting, backups, and ops
  • Embedding every chunk, re-embedding when models change
  • A reranker to make raw results usable
  • Dedup and reconciliation code nobody wants to own
  • Engineering time, forever
  • 1,627 tokens of "relevant" chunks on every single query

The token line is the one that scales. In our LoCoMo runs, recall scored 88.7% to raw search's 88.2% while returning 1,384 fewer tokens per query. As an illustration, at $2 per million input tokens that is over $2,700 per million queries off your model bill. Your own saving depends on your model's price and tokenizer.

The study behind the chat.

Recalld Chat runs on the same recall engine we benchmarked on LoCoMo (1,540 questions): 88.7% answer accuracy at 243 tokens per query, measured by us on 6 August 2026 on a third-party harness we didn't write, with its prompts and judge untouched. Adapters, patches, raw reports and known limitations are in our results repo.

88.7% accuracy 243 tokens / query 6.7× smaller context Third-party harness

One balance. Chat included.

Chat runs on the same credits as the memory API: model usage, web search, and memory all draw from one balance you can read.

Free
$0
15k credits / month
  • Default model
  • 30-day active memory
  • 60 req/min
  • Community support
Starter
$10 /mo
200k credits / month
  • Default model
  • 6-month active memory
  • 120 req/min
  • Email support · CSV export
Scale
$99 /mo
3.96M credits / month
  • All models · best $/credit
  • Unlimited active memory
  • 1,200 req/min · BYOK
  • Advanced analytics · Priority support

Need one-off credits? Top-ups from $5. All plans include the full API and Chat.

Your next chat won't start from zero.

Recall, not replay. Start free in the EU or US region today.