Dashboard

The dashboard is where you manage a FishMem project: inspect what the API is doing, browse stored memory, try things interactively, and handle keys and billing. Here's what each section is for.

Overview

The landing view for a project. It summarizes activity at a glance — request volume, the entities you're storing memory for, and overall memory activity — so you can see the project's pulse without digging in.

Requests

Inspect metered API activity request by request. Use this to confirm calls are landing, see which operations ran, and trace what your integration actually sent when something looks off. Each metered call shows up here.

Entities

Browse the users, agents, and runs that have stored memory. These are the user_id, agent_id, and run_id scopes from your API calls. Drill into an entity to see the memory held against it — useful for confirming a given user's memories are landing in the right scope.

Entities are computed from active canonical memories rather than maintained in a separate table. A record with user, agent, and run scope appears in all three aggregates. Removing an entity removes its matching memories from active recall and records deletion tombstones; it does not delete a named person or concept from the graph as an independent object.

Memories

Browse, search, and curate the exact stored memories. Search uses the same recall path as the API. Review provenance, correct inaccurate entries, and delete information that should no longer be retained.

Sources

Upload PDF, Office, EPUB, email, image, or text files and watch durable extraction move from queued to processing to ready. You can also paste already-textual content for synchronous ingest. Search citation-ready chunks, inspect the indexed Markdown and SHA-256 provenance, and permanently delete a complete source family.

Reusing a stable source key creates an immutable new version and moves the current head. File upload accepts up to 25 MB and 300 extracted pages; direct text accepts up to 1 MB. The raw file and extraction artifacts are removed with the source family or project.

Playground

An interactive harness for the core loop: add an authoritative memory, then recall it with a query. It is the fastest way to validate scope and ranking before writing integration code.

API keys

Create, rotate, and revoke project-scoped fm_... keys. New keys are shown once and stored hashed at rest. See Authentication for how to use and rotate them safely.

Usage and billing

Track credit consumption against your plan and manage your subscription. Billing runs through the Stripe customer portal, reachable from here. See Usage and credits and Billing for details.

Operations

Inspect durable memory inference, document extraction, export, import, namespace purge, restore, profile, and projection-rebuild work. A memory inference row shows its Event ID, live status, attempts, terminal error, and refined-record count without exposing the original conversation. Running and failed tasks remain visible across process restarts and failed tasks can be queued for a bounded retry.

Webhooks

Configure signed webhooks per project for memory mutations and source ingest/deletion so you can sync them into your own systems. See Webhooks.

Settings

Project-level configuration and account settings live here. The Backup tab downloads a versioned canonical snapshot or CSV subset and restores a JSON snapshot into an empty project. JSON snapshots include canonical memories, journal/history, exact document source versions, current heads, and deterministic chunks; derived vectors are rebuilt. Restore preserves IDs and is tracked in Operations.

Deleting a project permanently purges its canonical records, vectors, derived state, journal, queued work, keys, webhooks, and operational metadata. Cloud blocks deletion while a Hosted usage reservation is in flight, and a paid project must have its subscription canceled first.