This adds an ASSISTANT_CHAT_COMPLETION_ENDPOINT which can be used
to enable AI Assistance instead of an OpenAI API key. The assistant
then works against compatible endpoints, in the mechanical sense.
Quality of course will depend on the model. I found some tweaks
to the prompt that work well both for Llama-2 and for OpenAI's models,
but I'm not including them here because they would conflict with some
prompt changes that are already in the works.
Co-authored-by: Alex Hall <alex.mojaki@gmail.com>
Summary:
{F74398}
Refactored the 'radio checkboxes' in the modal for deleting a page and reused them here.
The option to download as a template already existed in the server code but wasn't being exercised by the frontend. Also added an option to remove just the history, which is the main motivation for this diff.
Test Plan: Expanded the existing nbrowser test.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3999
Summary:
Sometimes the model repeats the classes given in the prompt which would mess up extracting the actual formula. This diff solves this by:
1. Changes the generated Python schema so that (a) the thing that needs completing is a plain top level function instead of a property/method inside the class and (2) the classes are fully valid syntax, which makes it easier to
2. Remove classes from the parsed Python code when converting the completion to a formula.
3. Tweak the prompt wording to discourage including classes in general, especially because sometimes the model tries to solve the problem by defining extra methods/attributes/classes.
While I was at it, I changed type hints to use builtins (e.g. `list` instead of `List`) to prevent `from typing import List` which was happening sometimes and would look weird in a formula. Similarly I removed `@dataclass` since that also implies an import, and this also fits with the tweaked wording that the classes are fake.
Test Plan:
Added a new test case to the formula dataset which triggers the unwanted behaviour. The factors that seem to trigger the problem are (1) a small schema so the classes are easier to repeat and (2) the need to import modules, which the model wants to place before all other code. The case failed before this diff and succeeded after. The tweaked wording reduces the chances of repeating the classes but didn't eliminate it, so forcibly removing the classes in Python was needed.
There were also a couple of other existing cases where repeating the classes was observed before but not after.
Overall the score increased from 49 to 51 out of 69 (including the new case). At one point the score was 53, but changes in whitespace were enough to make it drop again.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D4000
Summary:
A floating formula editor is available by default and in the basic setup allows just formula modification.
AI assistant is now an optional component of the floating editor and it is controlled by OPENAPI_KEY presence.
Env variable GRIST_FORMULA_ASSISTANT was removed, new feature flag HAS_FORMULA_ASSISTANT is derived from the presence of OPENAPI_KEY.
Also updated anonymous signup nudge. By default it displays only info that this feature is only for logged in users.
Test Plan: updated
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3987
Summary:
- Implements MemoryPool for waiting on memory reservations.
- Uses MemoryPool to control memory used for stringifying JSON responses in Client.ts
- Limits total size of _missedMessages that may be queued for a particular client.
- Upgrades ws library, which may reduce memory usage, and allows pausing the websocket for testing.
- The upgrade changed subtle behavior corners, requiring various fixes to code and tests.
- dos.ts:
- Includes Paul's fixes and updates to the dos.ts script for manual stress-testing.
- Logging tweaks, to avoid excessive dumps on uncaughtError, and include timestamps.
Test Plan:
- Includes a test that measures heap size, and fails without memory management.
- Includes a unittest for MemoryPool
- Some cleanup and additions to TestServer helper; in particular adds makeUserApi() helper used in multiple tests.
- Some fixes related to ws upgrade.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3974
Summary:
The formula that's used when the Apply button is clicked, and the formula that's
shown in responses from the Formula Assistant should now be the same. Previously, they
would differ slightly.
Test Plan: Server tests.
Reviewers: alexmojaki
Reviewed By: alexmojaki
Subscribers: alexmojaki
Differential Revision: https://phab.getgrist.com/D3977
* move getTemplateOrg method; enable template org in docker tests
This moves the `getTemplateOrg` method to a neutral venue for the
convenience of `grist-static`, otherwise a lot of awkward dependencies
get pulled in needlessly in new parts of the app.
This also fixes docker tests using the template org.
* REQUEST now supports POST
* Add extra flag for enabling REQUEST, also update README and comments
Co-authored-by: John Cant <a.jonncant@gmail.com>
Co-authored-by: Alex Hall <alex.mojaki@gmail.com>
Summary:
The changes are intended to smooth over some sharp edges when a signed-out user
is using Grist (particularly while on the templates site).
Test Plan: Browser tests.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3957
Summary:
Fixes a problem reported here: https://community.getgrist.com/t/exporting-the-records-in-a-linked-view/2556/4
The download CSV/Excel link now contains an additional `linkingFilter` URL parameter containing JSON-encoded `filters` and `operations`. This object is originally created in the frontend in `LinkingState`, and previously it was only used internally in the frontend. It would make its way via `QuerySetManager` to `QuerySet.getFilterFunc` where the actual filtering logic happened. Now most of that logic has been moved to a similar function in `common`. The new function works with a new interface `ColumnGettersByColId` which abstract over the different ways data is accessed in the client and server in this context. There's no significant new logic in the diff, just refactoring and wiring.
Test Plan: Expanded two `nbrowser/SelectBy*.ts` test suites to also check the contents of a downloaded CSV in different linking scenarios.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3961
Summary:
Replaces https://phab.getgrist.com/D3940, particularly to avoid doing potentially unwanted things automatically.
Adds optional fields `evaluateCurrentFormula?: boolean; rowId?: number` to `FormulaAssistanceContext` (part of `AssistanceRequest`). When `evaluateCurrentFormula` is `true`, calls a new function `evaluate_formula` in the sandbox which computes the existing formula in the column (regardless of anything the AI may have suggested) and uses that to generate an additional system message which is added before the user's message. In theory this could be used in an interface where users ask why a formula doesn't work, including possibly a formula suggested by the AI. For now, it's only used in `runCompletion_impl.ts` for experimenting.
Also cleaned up a bit, removing `_chatMode` which is always `true` now, and uses of `regenerate` which is always `false`.
Test Plan: Updated `runCompletion_impl` to optionally use the new feature, in which case it now scores 51/68 instead of 49/68.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3970
Summary: Also fixes a few bugs with some telemetry events not being recorded.
Test Plan: Manual.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3960
* Fix support of Scaleway S3 bucket #359
While MinIO and AWS return versionId as strings, other S3 API
implementations return versionId as integers.
We must carefully convert the versionId as string in order to cover
these various behaviors.
Also ensure that docStorage is initialized before attempting to
calculate the data size in order to avoid an exception.
* Add unit tests for MinIOExternalStorage#versions() #359
Introduced some unit tests to :
- ensure listObjects is called with the right arguments;
- cover the case when a S3 bucket implementation does not return the
versionId as a string but rather as an integer (like Scaleway):
in such a case, ensure that the returned snapshotId is a string;
- cover the case when the listObjects function emits an error, ensure the
versions() call rejets with the error emitted;
- that the deleteMarkers are only returned when the
includeDeleteMarkers is passed;
---------
Co-authored-by: Florent FAYOLLE <florent.fayolle@beta.gouv.fr>
Summary:
- Replace logger module by the standard module 'logging'.
- When a log message from the sandbox includes newlines (e.g. for tracebacks),
keep those lines together in the Node log message.
Previously each line was a different message, making it difficult to view
tracebacks, particularly in prod where each line becomes a separate message
object.
- Fix assorted lint errors.
Test Plan: Added a test for the log-line splitting and escaping logic.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3956
Summary:
In a nutshell:
- More specific and helpful error messages are shown to the user
- API requests are only retried when needed
- The system deals with reaching the maximum token limit better, especially by switching to a model with a bigger limit
In more detail:
- `COMPLETION_MODEL` configuration has been removed. By default `gpt-3.5-turbo-0613` is used which accepts 4k tokens. If that's not enough, `gpt-3.5-turbo-16k-0613` is used instead.
- Switching to the bigger model happens when either the prompt is too long by itself (the API immediately returns an error code) or the model reaches the 4k limit itself in the process of generating a response and thus returns an incomplete response. The latter case is made possible by removing the `max_tokens: 1500` in the request, which was very generous and would have lead to switching to the more expensive model more often than needed. The downside is that the user has to wait a bit longer for the response.
- If the bigger 16k token limit is also exceeded, the assistant immediately responds (instead of retrying as before) with an error message including suggestions. The suggestions include restarting the conversation if and only if the user has sent multiple messages.
- If a request fails because Grist has reached its OpenAI monthly billing quota, the assistant immediately responds (instead of retrying as before) with an error message suggesting that the user try again tomorrow.
- If a request fails for some other reason, the assistant retries, and if all attempts fail then the user is told to try again in a few minutes and is shown the exact error message, including the API response if there is one.
- Retrying only happens when an API request fails, whereas previously the system also retried errors from a much bigger scope which included calls to the sandbox. The downside is that the hugging face assistant no longer retries, although that code is currently disabled anyway.
- The assistant no longer waits an additional second after the final retry attempt fails.
Test Plan: Added a new server test file with several unit tests using faked OpenAI responses, including the happy path which wasn't really tested before.
Reviewers: dsagal
Reviewed By: dsagal
Subscribers: dsagal
Differential Revision: https://phab.getgrist.com/D3955
Summary: Using standard tost notification, message about webhook queue being overflown was added. message is permanent as long as queue is full. Message contains linkt to the webhook setings
Test Plan: two nbrowser test was added - one to check if message is show when queue is full, and second to check if message is dismiss when queue was cleaned.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Subscribers: jarek
Differential Revision: https://phab.getgrist.com/D3929
Summary:
Remove a duplicate import, perhaps introduced during merge of
a relatively long-lived branch.
Test Plan: existing tests should pass
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3953
Summary:
introduces POST /api/docs/{docId}/webhooks and DELETE /api/docs/{docId}/webhooks/{webhookId} on place of old _subscribe and _unsubscribe endpoints.
Remove checking for unsubscribeKey while deleting webhook - only owner can delete webhook using DELETE endpoint. subscription key is still needed for _unsubscribe endpoint.
old _unsubscribe and _subscribe endpoints are still active and work as before - no changes there.
Posting schema:
```
POST /api/docs/[docId]/webhooks
```
Request Body:
```
{
"webhooks": [
{
"fields": {
"url": "https://webhook.site/3bd02246-f122-445e-ba7f-bf5ea5bb6eb1",
"eventTypes": [
"add",
"update"
],
"enabled": true,
"name": "WebhookName",
"memo": "just a text",
"tableId": "Table1"
}
},
{
"fields": {
"url": "https://webhook.site/3bd02246-f122-445e-ba7f-bf5ea5bb6eb2",
"eventTypes": [
"add",
],
"enabled": true,
"name": "OtherWebhookName",
"memo": "just a text",
"tableId": "Table1"
}
}
]
}
```
Expected response: WebhookId for each webhook posted:
```
{
"webhooks": [
{
"id": "85c77108-f1e1-4217-a50d-acd1c5996da2"
},
{
"id": "d87a6402-cfd7-4822-878c-657308fcc8c3"
}
]
}
```
Deleting webhooks:
```
DELETE api/docs/[docId]/webhooks/[webhookId]
```
there is no payload in DELETE request. Therefore only one webhook can be deleted at once
Response:
```
{
"success": true
}
```
Test Plan: Old unit test improved to handle new endpoints, and one more added to check if endpoints are in fact created/removed
Reviewers: alexmojaki
Reviewed By: alexmojaki
Subscribers: paulfitz, alexmojaki
Differential Revision: https://phab.getgrist.com/D3916
Grist by default uses node-sqlite3 to manipulate data in an
SQLite database. If a single parameter is passed to `run`
and it is a list, the list is unpacked and its contents treated
as the actual parameters. In grist-static, we use other SQLite
interfaces that don't have that automatic unpacking. Most
calls like this have been removed from Grist, but at least one
was missed, and was causing symptoms such as
https://github.com/gristlabs/grist-static/issues/5
This change should make no difference to regular Grist, but
resolves the grist-static problems.
Summary:
add back config call for external storage
in the right order.
Test Plan: existing saas tests catch this
Reviewers: dsagal
Reviewed By: dsagal
Differential Revision: https://phab.getgrist.com/D3946
Summary:
Adding limits for AI calls and connecting those limits with a Stripe Account.
- New table in homedb called `limits`
- All calls to the AI are not routed through DocApi and measured.
- All products now contain a special key `assistantLimit`, with a default value 0
- Limit is reset every time the subscription has changed its period
- The billing page is updated with two new options that describe the AI plan
- There is a new popup that allows the user to upgrade to a higher plan
- Tiers are read directly from the Stripe product with a volume pricing model
Test Plan: Updated and added
Reviewers: georgegevoian, paulfitz
Reviewed By: georgegevoian
Subscribers: dsagal
Differential Revision: https://phab.getgrist.com/D3907
Summary:
In a selector table, when a selected row is filtered out of view, linked widgets should update based on the newly selected row.
There were a few bugs that contributed to this wrong behavior:
- Gridview wasn't subscribing to the current row id, and the row with id 'new' was being converted to the first row
- Cursor was keeping track of the currently selected row id, it was hiding a problem behind the proper rowIndex
- Undo/redo somehow leveraged the wrong rowId from the cursor during the position restore.
The `No data` text was also changed to be more meaningful.
Test Plan: Added and updated.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3937
Summary:
Adds a new Support Grist page (accessible only in grist-core), containing
options to opt in to telemetry and sponsor Grist Labs on GitHub.
A nudge is also shown in the doc menu, which can be collapsed or permanently
dismissed.
Test Plan: Browser and server tests.
Reviewers: paulfitz, dsagal
Reviewed By: paulfitz
Subscribers: jarek, dsagal
Differential Revision: https://phab.getgrist.com/D3926
Summary:
Here's a series of badness that easily leads to a crash, in reverse order:
- Lodash's map() function interprets an object with a .length property as an array.
- Some very old code generated human-friendly descriptions of user actions,
applying map() to parts of them. It so happens that this generated description
isn't even used.
- If a user action is encountered with a sufficiently large length propery,
map() would exhaust the server memory.
Fixed by removing old unneeded code, and replacing some other occurrences of
lodash's map() with native equivalents.
Test Plan: Tested manually on a local reproduction of the issue.
Reviewers: paulfitz
Reviewed By: paulfitz
Subscribers: paulfitz
Differential Revision: https://phab.getgrist.com/D3938
Summary:
This tweaks the prompting so that the user's message is given on its own instead of as a docstring within Python. This is so that the prompt makes sense when:
- the user asks a question such as "Can you write me a formula which does ...?" rather than describing their formula as a docstring would, or
- the user sends a message that doesn't ask for a formula at all (https://grist.slack.com/archives/C0234CPPXPA/p1687699944315069?thread_ts=1687698078.832209&cid=C0234CPPXPA)
Also added wording for the model to refuse when the user asks for something that the model cannot do.
Because the code (and maybe in some cases the model) for non-ChatGPT models relies on the prompt consisting entirely of Python code produced by the data engine (which no longer contains the user's message) those code paths have been disabled for now. Updating them now seems like undesirable drag, I think it'd be better to revisit this when iteration/experimentation has slowed down and stabilised.
Test Plan:
Added entries to the formula dataset where the response shouldn't contain a formula, indicated by the value `1` for the new column `no_formula`.
This is somewhat successful, as the model does refuse to help in some of the new test cases, but not all. Performance on existing entries also seems a bit worse, but it's hard to distinguish this from random noise. Hopefully this can be remedied in the future with more work, e.g. automatic followup messages containing example inputs and outputs.
Reviewers: paulfitz
Reviewed By: paulfitz
Subscribers: dsagal
Differential Revision: https://phab.getgrist.com/D3936
Summary:
It became hard to detect aborted connections in node 16.
In node 14, req.on('close', ...) did the job. Thid diff adds a
work-around, until a better way is discovered or added.
Aborting a req will typically lead to 'close' being called
on the response, without writableFinished being set.
- https://github.com/nodejs/node/issues/38924
- https://github.com/nodejs/node/issues/40775
Test Plan:
existing DocApiForwarder test passes; manually
checking on various node versions.
Reviewers: JakubSerafin
Reviewed By: JakubSerafin
Differential Revision: https://phab.getgrist.com/D3923
Summary:
The previous code for extracting a Python formula from the LLM completion involved some shaky string manipulation which this improves on.
Overall the 'test results' from `runCompletion` went from 37/47 to 45/47 for `gpt-3.5-turbo-0613`.
The biggest problem that motivated these changes was that it assumed that code was always inside a markdown code block
(i.e. triple backticks) and so if there was no block there was no code. But the completion often consists of *only* code
with no accompanying explanation or markdown. By parsing the completion in Python instead of JS,
we can easily check if the entire completion is valid Python syntax and accept it if it is.
I also noticed one failure resulting from the completion containing the full function (instead of just the body)
and necessary imports before that function instead of inside. The new parsing moves import inside.
Test Plan: Added a Python unit test
Reviewers: paulfitz
Reviewed By: paulfitz
Subscribers: paulfitz
Differential Revision: https://phab.getgrist.com/D3922
Summary:
Previously we failed to log signup info for users who signed up via
Google. This fixes that issue by recording it on first post-signup
visit. It also includes signup as a new telemetry event, recorded at the
same point.
Test Plan: Tested locally to see that a signup produces an appropriate log message and telemetry event.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3921
Summary:
- Move css module for the login page css to core/, to be reusable in core/ pages.
- Move /welcome/teams implementation to WelcomeSitePicker.ts
- List users for personal sites, as well as team sites.
- Add org param to setSessionActive() API method and end endpoint, to allow
switching the specified org to another user.
- Add a little safety to getOrgUrl() function.
Test Plan: Added a test case for the new behaviors of the /welcome/teams page.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3914
Summary:
This adds a `yarn cli settings telemetry [--json] [--all]` command
that allows telemetry settings to be inspected. It is useful for
keeping documentation about telemetry up to date.
Test Plan:
manual (a bit cheeky; justified on basis of breakage
not being very important yet, this is essentially an internal
feature)
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3917
Summary: Also fixes a few small bugs with telemetry collection.
Test Plan: Server and manual tests.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3915
Summary:
For grist-static, we want to the data engine to be able to call external/exported JS functions directly,
rather than via the node 'server' living in another thread which requires synchronous communication hackery.
As a step in that direction, this diff changes the exported functions that we care about (guessColInfo and convertFromColumn)
to just using the top-level functions instead of relying on fields in ActiveDoc, namely docData.
For guessColInfo, this is done by directly passing the small amount of metadata that was previously retrieved from the DocData.
For convertFromColumn, disentangling DocData is a lot more complicated, so instead we construct a fresh DocData object using
the required metadata tables which are now passed in by the data engine.
Test Plan: Existing tests
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3913
Summary:
Adds support for optional telemetry to grist-core.
A new environment variable, GRIST_TELEMETRY_LEVEL, controls the level of telemetry collected.
Test Plan: Server and unit tests.
Reviewers: paulfitz
Reviewed By: paulfitz
Subscribers: dsagal, anaisconce
Differential Revision: https://phab.getgrist.com/D3880
Summary:
sanitazing errors output in webhooks to protect users data (not show them in logs and other places).
Because redis is returing whole payload when error occur, best approach is to hijack exception as close to redis operation as posible and sanitize the data.
We need to know data structure do do this corretly tho. Currently I decided to just censore everything that has "payload" key.
Test Plan: Because logs that need to be sanitized come from redis, to be valid tested we should force redis to crash. It's hard to do in our integration test setup. In this moment, unit test is all we got.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3905
Summary:
Adding a way to detach an editor. Initially only implemented for the formula editor, includes redesign for the AI part.
- Initially, the detached editor is tight with the formula assistant and both are behind GRIST_FORMULA_ASSISTANT flag, but this can be relaxed
later on, as the detached editor can be used on its own.
- Detached editor is only supported in regular fields and on the creator panel. It is not supported yet for conditional styles, due to preview limitations.
- Old code for the assistant was removed completely, as it was only a temporary solution, but the AI conversation part was copied to the new one.
- Prompting was not modified in this diff, it will be included in the follow-up with more test cases.
Test Plan: Added only new tests; existing tests should pass.
Reviewers: JakubSerafin
Reviewed By: JakubSerafin
Differential Revision: https://phab.getgrist.com/D3863
Summary:
- Move makeXLSX* methods to workerExporter file to avoid the risk of creating a piscina worker pool from a thread.
- Increase request timeout in ExportsAccessRules test that started failing occasionally
Test Plan: Test should succeed more reliably
Reviewers: jarek
Reviewed By: jarek
Differential Revision: https://phab.getgrist.com/D3910
Paul Fitzpatrick
Alex Hall
Florent
Jarosław Sadziński
Dmitry S
George Gevoian
John Cant
Jakub Serafin
Florent