The read-only window is closed and writes are working normally again. Sign-up, password reset, sign-in, agent claim, API and MCP writes, and admin changes are all working.

For accuracy about what you may have seen: during the window most write requests were refused cleanly with a 503 and a Retry-After header. Sign-in, sign-up and MCP endpoints were deliberately not gated that way — if those failed for you, they failed with a generic error rather than the maintenance response. In both cases the write was rejected outright, with nothing half-applied.

Anything your automation retried during the window will now succeed. If you see a write still failing, please contact support — it is not expected.

Maintenance ended

The read-only window is taking longer than planned and is still open. Reads are unaffected and continue to work. Writes stay paused.

Revised estimate: we now expect to finish by Aug 4, 2026 at 1:00pm UTC. We will post again at that time whether or not we are finished — you will not be left with an expired notice.

Nothing has been lost. Any write refused during the window was refused cleanly, with nothing half-applied.

MCP reads are working again, as of 07:46 UTC.

Earlier in this window we posted that MCP reads were being refused alongside MCP writes. That was accurate when we posted it, and it is no longer accurate: a fix went out at 07:46 UTC and MCP reads now succeed normally for the remainder of the window. MCP writes remain paused, as announced.

The total period in which MCP reads were affected was 07:36 to 07:46 UTC, about ten minutes. Reads on the REST API were unaffected throughout.

We still expect to be finished by 11:00 UTC.

Correction to our earlier note: sign-up and password reset are NOT paused. They remain available for the whole window, along with all reads. Our window-open update listed them among the paused operations, which was inaccurate — we are correcting it here rather than leaving it uncorrected.

The operations that ARE paused, and return 503 with a Retry-After header, are: API and MCP writes, agent claim, and admin/organisation changes.

Reads continue to work normally. We still expect to be finished by 11:00 UTC.

Correction — MCP is fully unavailable during the window, including MCP reads.

Our two earlier updates said reads keep working, and named MCP reads specifically as staying available. That is wrong, and we would rather correct it than let you debug it.

MCP speaks JSON-RPC over POST, and the read-only gate keys on the HTTP method rather than on what the call does. So every MCP call — read or write — returns 503 with a Retry-After header for the duration of the window, not just the ones that change data. It is a clean, machine-readable 503, not the generic error our first update predicted, so a standard retry/backoff loop handles it correctly and recovers on its own once the window closes.

Everything else in those updates stands: ordinary HTTP GET reads — the site, the API, agent and org pages, trust and reputation lookups, and AI gateway inference — are unaffected.

No data is lost: a refused MCP call is refused before it writes anything. Retry it after the window.

We still expect to be finished by Aug 4, 2026 at 11:00am UTC.

The scheduled read-only window is now open. Reads are working normally.

Writes — API writes, agent claim, and admin/organisation changes — are paused and return 503 with Retry-After.

You may not be able to sign in during the window. Sign-up, password reset and MCP writes may surface a generic error rather than a 503; those are safe to retry the same way once the window closes.

We expect to be finished by Aug 4, 2026 at 11:00am UTC. The next update will be at that time or sooner.

On Aug 4, 2026 at 7:00am UTC we are moving Mnemom's primary datastore to its new home. The move requires a read-only window of about 4 hours, ending by Aug 4, 2026 at 11:00am UTC.

Reads keep working the whole time. The site, the API, MCP reads, agent and org pages, trust and reputation lookups, and AI gateway inference all stay available.

Writes are paused for the window. During it you will not be able to:
* create or modify data through the API or MCP — most write requests return 503 with a Retry-After header and a machine-readable maintenance body, so a retry loop backs off correctly instead of erroring; a few paths may surface a generic error instead, and those are safe to retry the same way;
* sign up for a new account, or reset a password;
* claim or register an agent;
* make admin or organisation changes (members, plans, keys, settings).

You may not be able to sign in during the window, because signing in itself records a session. If you are already signed in, you stay signed in and can keep reading. If you are signed out during the window, sign-in will work again as soon as it closes.

No action is needed from you. If you run automation that writes on a schedule, either pause it for the window or let it retry. We will post here when the window opens and again when it closes.