Cost control: a monthly token allowance for each modelSee how it works
memtro
Start for free

Replicate + Memtro

Which predictions failed today, and what is each deployment running on?

Replicate connects Memtro to your hosted AI models. It reads the account's predictions and trainings with status, timings, compute seconds and errors, never their inputs, outputs or logs. It reads deployments with hardware and instance counts, the hardware on offer, and public models. When asked, and once an administrator has granted the action by name, it can run a model or a deployment, which Replicate bills, cancel a run, start a fine-tune, create, resize or delete a deployment, create, describe or delete a model, and delete a model version. Replicate tokens have every right of their account.

Connect ReplicateStart for freeReference
In Memtro

What you can do with Replicate in Memtro

Ask in plain English from chat, your editor, a scheduled job or your own agent. The model picks the right Replicate tool, reads what it needs, and answers with the source cited.

What it does

  1. 01Lists predictions, most recent first, with model, version, status, timings, compute seconds and error, never inputs, outputs or logs
  2. 02Lists trainings with base model, status, timings, compute seconds and the version each produced
  3. 03Lists deployments with model, version, release, hardware and minimum and maximum instances, and the hardware Replicate offers
  4. 04Reads a model's description, visibility, run count, latest version and the inputs it takes, and searches public models by words
  5. 05When asked and granted, runs a model or a deployment, which Replicate bills per second or per output, starts a training, often costly, and cancels a running prediction or training
  6. 06When asked and granted, creates, changes or deletes a deployment, which changes its cost and what callers get, creates or describes a model, and permanently deletes a model or a model version with its predictions and output files

Acts when asked

When you ask it to

These tools change something in Replicate. The model is told to use them only when you ask for exactly that, and an administrator grants each one by name to the teams that may use it; a personal connection acts only after its owner allows it.

  1. 01Run a model (billed per second or per output by Replicate)
  2. 02Run a deployment (billed, and may wake an instance)
  3. 03Cancel a running prediction
  4. 04Start a training (fine-tune) of a model version into a destination model (billed)
  5. 05Cancel a running training
  6. 06Create a deployment of a model version on chosen hardware (billed while instances run)
  7. 07Change a deployment's hardware, instance counts or version
  8. 08Delete a deployment (only after 15 minutes offline)
  9. 09Create a model (public or private)
  10. 10Change a model's description, readme or links
  11. 11Delete a private model with no versions
  12. 12Delete a model version, with all its predictions and output files

Events

When it happens

Turn events on in the connection's Events tab on the Connections page, then add the address it shows as a webhook in Replicate, choosing the same kinds; the tab shows Replicate's own steps. Each event can start a job, with the event as its input. Replicate signs every delivery, and one whose signature does not check out is refused before it is read.

Events and webhooks
  • 01A run is starting
  • 02A run is processing (new output or logs)
  • 03A run succeeded
  • 04A run failed
  • 05A run was cancelled
  • 06A run was aborted before it started
In practice

Ask it like you would a colleague

Once Replicate is connected, these are the kinds of things people ask. Plain questions, no syntax. If the answer lives partly in another system you have connected, Memtro looks there too.

Replicate + Memtro

  • Which predictions failed today?

  • How many compute seconds did yesterday's runs use?

  • What hardware is the image deployment running on?

  • What inputs does this model take?

  • Scale the image deployment down to zero minimum instances

01

Failures after a release

Maya Okafor, ML engineer at Acme Ltd, asked which predictions had failed since the morning release. Memtro listed them with model, version, status and error, and showed every failure came from one new version.

02

Cutting idle cost

Tom Reilly, platform lead at BigCo, asked Memtro to set the image deployment's minimum instances to zero. Memtro changed it in Replicate, taking effect as a new release at once, so the hardware is no longer billed while idle.

Teams

Where it fits

The teams that lean on Replicate most, a question each of them asks, and what they connect beside it.

Better together

Works with

A record in Replicate is rarely the whole story. Connect the systems that hold the rest and Memtro checks them together, then remembers what it learned so the next question starts further along.

Also in Engineering and operations

Using something else for the same job, or both? Each connector has its own tools; connect as many as you like.

Setup

How to connect

Connect withAPI token
  1. 1

    In Replicate, open Account → API tokens and choose the user or organization Memtro should act for.

  2. 2

    Create a token named Memtro and copy it. It starts r8_ and has every right of its account: runs Memtro starts are billed there, only when somebody asks.

  3. 3

    In Memtro, open Connections → Replicate and paste it into API token.

  4. 4

    Ask your first question: which predictions failed today?

Also works from

Chat in the dashboard, the OpenAI-compatible API, scheduled jobs, and any MCP client: Claude Code, OpenCode, Cursor, Claude Desktop. The Replicate tools are the same everywhere.

Set it up as an MCP serverClients and agents
FAQ

Questions

Can Memtro change anything in Replicate?+

Only when you ask, and only what it has been granted: run a model (billed per second or per output by Replicate); run a deployment (billed, and may wake an instance); cancel a running prediction; start a training (fine-tune) of a model version into a destination model (billed); cancel a running training; create a deployment of a model version on chosen hardware (billed while instances run); change a deployment's hardware, instance counts or version; delete a deployment (only after 15 minutes offline); create a model (public or private); change a model's description, readme or links; delete a private model with no versions; delete a model version, with all its predictions and output files. Every other tool reads, searches or lists. The Connections page labels the tools that act, an administrator grants each one by name to the teams that may use it, and a personal connection acts only after its owner allows it.

Who can use a Replicate connection?+

A personal connection serves only your own requests. A shared connection serves the organisation: everyone, or only the teams an admin chooses. Admins can always add shared connections; other members need the permission.

Does Memtro store my Replicate data?+

Credentials are encrypted at rest. Tool results are cached briefly so agent loops do not repeat identical calls, and the facts the model extracts are saved to your organisation's memory with provenance (system, record and link) so they can be checked and superseded.

How do I connect Replicate?+

API token. In Replicate, open Account → API tokens and choose the user or organization Memtro should act for.

Can Replicate start a job?+

Yes. Turn events on in the connection's Events tab on the Connections page, then add the address it shows as a webhook in Replicate, choosing the same kinds; the tab shows Replicate's own steps. A job set to start "When an event arrives" then runs once for each matching event, with the event as its input. The run treats the event as outside data, never as instructions, and anything the run remembers is kept as unverified.

Connect Replicate in a few minutes

Seven-day free trial, then one monthly price per organisation. Connectors, jobs and memory are never metered.

Start for freePricing

Where agents remember your company

Sign in, add a provider key, create an API key and point your tools at Memtro. It takes about five minutes.

Start for freeGetting started

7-day free trial · card on file, nothing charged until it ends · cancel any time