Chronos/DBWaitlist

Changelog

Chronos Gateway

All 24 releases, newest first, each with its full notes. Downloads · Docs · Chronos DB changelog

0.4.4

With Chronos DB 0.6.211 changes

  • The plain-words diff of a run never shows a masked column's value. Its changes (an approval link, the console, GET /v1/runs/:id/changes) read (masked) there, as the undo preview already did; the engine's admin diff shows them.
  • A deleted webhook gets nothing more. Deliveries still due to it were sent on through their retries, about 8.6 hours, approval links in Slack messages included; they are dropped now.
  • A webhook's URL is never logged, and its secret is kept sealed. A Slack webhook's URL is its credential: the give-up warning names the webhook's id, and the error kept with a delivery leaves the URL out. Secrets made before are read as they are.
  • One receiver that doesn't answer no longer holds the others back. A webhook or an effects channel whose receiver timed out (or refused the connection) is tried once a pass; its other deliveries wait for the next, so Slack approvals and refunds on other channels go out meanwhile.
  • A Chronos Cloud undo of a run is recorded in the run, as through the API: the run says undone, merge.undone goes to the webhooks, and the run's merges are undone together (a part of them, 400).
  • MongoDB tools write all of a call's documents or none. mongo_insert and mongo_update_many check every document before the first is written: a bad one refused the call after the ones before it were written, and a retry wrote those twice.
  • MongoDB updates work beside a masked ObjectId or integer field. A lifted field the update leaves alone is neither checked nor written; one the run read masked (as text) failed every update of the collection.
  • A MongoDB tool call counts its writes against GATEWAY_RUN_STATEMENTS_PER_MINUTE, before the first: one call counted once however many documents it wrote. A call with more writes than the limit left is refused whole (429).
  • A new setup code works after a lockout. The count of wrong setup codes is kept with the code, so one made by chronos-gateway setup-code (another process) starts again, as the error says; before, only a restart did.
  • deploy/compose.yaml pulls latest by default, not 0.2.11, whose image can't start its engine.
  • The run list, status counts and sweeper use indexes, and webhook deliveries done with for 30 days are deleted once an hour: each of these read every run, under the store's one lock, at every call.

0.4.3

With Chronos DB 0.6.26 changes

  • A merge from Chronos Cloud is a person's approval of a run, as through the API. For a run's world it is refused unless the run is held or in conflict (409, the run's status as the code), unless it names the version the person saw (400), and with confirm or into (400). Before, the cloud could merge a run its agent was still writing, or one written to after the person looked. A world that is not a run's merges as before.
  • Masked and hidden columns no longer reach the cloud in diffs mode by other ways. A cloud diff never returns SQL (sql: true is dropped: it quoted every column), and a refusal loses its detail and, for an integrity error, its message, which quote the rows, as counts mode already did.
  • A late approval that meets a conflict is kept until the hold ends. A held run approved past its expires whose merge met a conflict was expired by the next sweep, its world dropped; a run in conflict now waits until hold_expires, as a held one does.
  • The gateway listens while its engines start and a first attach copies the database. Without a shared store, it used to listen only after them, so /health was shut for a long first attach and an orchestrator's liveness probe killed it, again on every start. Requests that need the engine meanwhile answer 502.
  • MongoDB $exists: 0 and $exists: null ask for documents without the field, as MongoDB reads them. They were read as $exists: true, so mongo_delete_many {zip: {$exists: 0}} deleted the documents that had zip.
  • A console sign-in made with GATEWAY_API_KEY ends when that key is changed. It was checked only for some env key being set, so a sign-in made with a leaked key outlived its rotation, up to 12 hours. Sign-ins made with it before this release end at the upgrade: sign in again.

0.4.2

With Chronos DB 0.6.21 change

  • A strict change is superseded by a commit to the database since its hold, as the engine's strict proposals are. Its hold records strict.target (the stamp of the world it merges into, from a preview of the merge) and every approval of it, through the API, a link or the cloud, merges bound to it: one approved after other changes reached the database is refused 409 strict_target and nothing is merged (reject it; the agent proposes again). A strict change is approved whole: only_tables/only_keys on one are refused 400.

0.4.1

With Chronos DB 0.6.22 changes

  • Engine 0.6.2, which this release bundles: strict proposals for payments (only a person approves one, within 15 minutes, and a commit to main since supersedes it), merges into an attached Postgres or MySQL refused when it may lose a commit it acknowledged, MongoDB merges committed to a majority and journaled, and fixes from an audit of 0.6.0. The data format is unchanged (8).
  • Strict changes are for a person, in their window. A held run whose merge sends payment, payout, transfer or refund effects (or ones its merge policy names in strict_channels; engine 0.6.2) records strict: {channels, until} in its hold, merge.held and the propose's answer. The AI reviewer never approves one (it passes it to a person with why); a person's approval after until is refused 409 window_passed; GET /v1/approve returns strict and <chronos-approve> shows it with the time left, its Approve button off once it runs out.

0.4.0

With Chronos DB 0.6.07 changes

  • Engine 0.6.0, which this release bundles: an agent held to a tenant is kept to its tenant's worlds and rows everywhere, every merge into main is logged with the rows it wrote in a hash-chained ledger (VERIFY LEDGER), and merges check the constraints they bring against rows the run never saw. Upgrading: the engine's first checkpoint or merge writes data format 8, which 0.5 can't open: back up the data folder before you upgrade.
  • Runs are safe against lost answers and two decisions at once. A run approved by the AI reviewer or through Chronos Cloud keeps its merge id, so it can be undone like one approved through the API. An approval always names the version it merges (the one the run was held at when the caller names none), never whatever the world holds by then. A run's status moves only from what it was read as: a person's approve, the reviewer, the cloud and the sweeper no longer overwrite one another (the second gets 409 run_changed). When a propose or approve times out, or the sweeper finds a run's world gone, the gateway asks the engine's history whether it merged and records it merged (reconciled: true) instead of expired; it checks live runs the same way when it starts.
  • Child runs. POST /v1/runs/:id/children delegates part of a run to a sub-agent: its own token, its world forked from the parent's, held to the parent's tenant and to no more than the parent may change. Its propose merges into the parent's world; the parent proposes once its children are done, and discarding it discards them. GET /v1/runs/:id lists children.
  • Effects delivered. With GATEWAY_EFFECTS (a receiver URL per channel) and GATEWAY_EFFECTS_SECRET, the gateway takes what merges queue in the engine's outbox (NOTIFY ON MERGE) and delivers each, signed, with its id as Idempotency-Key; retried with backoff, given up after GATEWAY_EFFECTS_MAX_ATTEMPTS (8). GET /v1/effects shows each one's status, attempts and receipt.
  • Evidence with a propose. POST /v1/runs/:id/propose takes evidence: [{kind, uri, sha256, note}], kept with the run (and attached to its world where the engine supports it). PATCH /v1/runs/:id/meta adds keys to a run's metadata.
  • Clients 0.3.0. Python: Run.undo(), Run.child(), Run.children(), Run.set_meta(), propose(evidence=), Gateway.effects(); adapters for CrewAI (crewai_tools), Pydantic AI (pydantic_ai_toolset, pydantic_ai_deps) and a LangGraph checkpointer (ChronosCheckpointSaver) that records each checkpoint on the run. TypeScript: run.undo(), run.child(), run.children(), gateway.effects(), propose({evidence}).
  • A merge the engine made is never recorded as rejected. A rejection or discard that finds the run's world already gone asks the engine's history first, and records the run merged when its merge landed (a rejection then gets 409 run_changed); an approval whose merge succeeded while a rejection or discard was recorded records it merged all the same (recorded_over in its outcome). Child runs count against GATEWAY_RUNS_PER_HOUR and are refused past it (429 too_many_runs) like any run. An effect whose channel has no receiver waits without using its delivery attempts. The reviewer's verdict no longer resets the version a part approval left. Python: the LangGraph checkpointer logs, and never raises, when the gateway can't be reached.

0.3.4

With Chronos DB 0.5.51 change

  • Deeply nested SQL runs again, and reads are faster. Engine 0.5.5, which this release bundles, runs statements nested 100 parentheses or function calls deep (release builds refused 80), and types a nested coalesce, greatest or least once a level (22 deep took 3 seconds, doubling with each level). Engine 0.5.4 made scans, GROUP BY and one-key joins 20 to 25% faster. This is what an agent's own SQL in a run meets; merges and undo are as in 0.3.3. The gateway's own code is unchanged.

0.3.3

With Chronos DB 0.5.31 change

  • Undo of a big run is about 20 times faster. The engine checked each row an undo puts back by reading the whole table; it finds the row by its key now (engine 0.5.3, which this release bundles). Undoing a run that wrote 10,000 rows of a 20,000-row table beside Postgres: 57 s before, 2.7 s now; the preview 52 s and 2.8 s. The gateway's own code is unchanged.

0.3.2

With Chronos DB 0.5.27 changes

  • Undo after a partial undo. A partly_undone run can be undone again: rows the first undo put back aren't counted as changed, and once the rows left hold what the run wrote, they go back and the run becomes undone. The console's button reads "Undo the rest". With an engine before 0.5.2 the second undo needs skip_changed.
  • Undo after a part was approved. A held run with a part approved (only_tables, only_keys) and the rest rejected, expired or discarded kept the part's rows in the database with no way to undo them through the gateway. POST /v1/runs/:id/undo and the console's Undo take such a run now.
  • PUT /v1/policy answers ended: the rules that were set and the new set left out. The API doc's example sent a part of the rules, which ends max_deletes = 0; it sends them all now and says so.
  • A Slack message is made once per delivery: a retry sends the same approval or undo link instead of making a new one each time (while the gateway runs; a restart makes one more).
  • Slack messages name MariaDB as MariaDB, not MySQL.
  • mongo_update_many and mongo_delete_many say in their descriptions that a filter matching more than 1,000 documents is refused, not cut to the first 1,000.
  • Docs: a write Postgres refuses under act_as fails the run (run.failed), not merge.rejected; approved_by.by from a developer key is the caller's word, unchecked.

0.3.1

With Chronos DB 0.5.17 changes

  • GATEWAY_FOLLOW beside MongoDB (engine 0.5.1 or later): main takes MongoDB's changes about a second after they commit, through a change stream on each attached collection. The role the connection check gives has changeStream on those collections for it. With several databases, --follow is given only to engines of databases that are followed (not MySQL or MariaDB yet).
  • MongoDB filters: $regex and dates. The MongoDB tools (find, count, update_*, delete_*) take {"$regex": ..., "$options": "i"} on strings and arrays of strings, and compare dates ({"$date": ...}) with $eq $ne $gt $gte $lt $lte $in $nin as the instants they are. Both were refused by name before.
  • Undo from Slack. A Slack webhook's message of a merge carries a link, Review and undo: a page that shows what would go back and undoes that merge, once, until GATEWAY_APPROVAL_TTL (GATEWAY_SLACK_UNDO=off: no link; with single sign-on, a link to the run in the console). The undo is the checked one: rows changed since stay. Slack messages name the database they are about (MySQL, MongoDB), and say when a merge is undone.
  • Disconnect a database. DELETE /v1/sources/:name (and Disconnect on the console's Connect page) takes away a database connected under a name: its engine stops, the gateway's copy and settings of it go, its runs stay as history, and the database itself is left as it is. Runs still in progress there are refused (409 runs_live) or discarded with it (?discard_runs=true). A run that is over still reads after its database is disconnected.
  • Undo with the gateway's own role. The statements the connection check gives for the gateway's role (role_sql) now make and grant the table that keeps what merges write (chronos.merge_rows; chronos_merge_rows beside MySQL and MariaDB). A role made from the 0.3.0 statements or earlier has none, so undo wasn't ready there (merges were never affected). POST /v1/runs/:id/undo now answers 409 undo_not_ready with the statements to run once, in place of the database's own error; GET /v1/status says attached.undo.ready.
  • Several gateways starting at the same moment on an empty Postgres store (pods of one set) could fail once making its tables, and restart. They take turns now.
  • Console: pages name the database being looked at (MySQL, MongoDB) where they said Postgres; a conflict beside MySQL or MongoDB shows its rows to settle, as beside Postgres; a question left open (approve a delete) closes when you go to another page, where it could be answered about the wrong thing; undo on a database not set up for it shows the statements to run. The connection check beside MySQL no longer lists chronos_merge_rows as a table to attach.

0.3.0

With Chronos DB 0.5.012 changes

  • The image can start its engine again. In the 0.2.11 image the engine (/usr/local/bin/chronos) had mode 0700 and belonged to root, while the image runs as user 65532: the gateway started, and then POST /v1/connect answered running /usr/local/bin/chronos attach: Permission denied (os error 13). The image now sets both programs to 0755 whatever the build machine's files had, and the release build fails if either is anything else. The installer's binaries were never affected.
  • MySQL and MariaDB. POST /v1/connect/check and POST /v1/connect take a mysql:// or mariadb:// URL (MySQL 8.0 or later, MariaDB 10.5 or later), and runs then work as beside Postgres: a world per run holding one tenant's rows, the rules, approvals, and a write to the database in one transaction with every row locked and compared. Agents write the same SQL (the engine's, Postgres's dialect) whatever the database. Not yet beside MySQL: binary columns, act_as, and GATEWAY_FOLLOW. The check says each of these per table or as a warning. The console's connect step takes either URL.
  • Undo a merged run. POST /v1/runs/:id/undo puts back every row the run wrote, as it was before, through the database as any merge goes: each row locked and compared, one transaction. preview: true says what would go back, in words, and which rows changed since; skip_changed: true leaves those and undoes the rest. A run becomes undone or partly_undone, and merge.undone or merge.undo_conflict is sent. A run keeps its merges' ids (merges). Beside Postgres, MySQL and MariaDB, for 30 days after the merge, with engine 0.5.0 or later. The preview names the tables the database would change besides the run's own (a cascade; Postgres only) and the effects that aren't undone, and an undo that would cascade needs allow_cascade. The console's run page has an Undo button: always the preview first, then one button. Not yet: an Undo button in Slack.
  • Runs add rows to AUTO_INCREMENT tables beside MySQL and MariaDB, with engine 0.4.12 or later: the row has a provisional id (negative) in the run, and MySQL's own once merged. The check's inserts is false only where the column is narrower than INT or is BIGINT UNSIGNED. Generated columns are attached (listed as generated), and a column with an expression default (DEFAULT CURRENT_TIMESTAMP) is no longer refused as one.
  • A run's reads are checked against MySQL and MariaDB, with engine 0.4.11 or later: a row a run read that changed in the database since holds its merge, as beside Postgres. The gateway asks the engine whether it can; with an older engine the default rules leave check_reads off, and PUT /v1/policy refuses to turn it on.
  • MongoDB. POST /v1/connect/check and POST /v1/connect take a mongodb:// URL (MongoDB 5.0 or later, a replica set), with an engine built with MongoDB (0.5.0 or later). A collection is attached as a table of _id, the fields named in fields (the tenant's among them), and the rest of each document. Runs work as beside Postgres: a world per run holding one tenant's documents, the rules, approvals, undo, and a write to MongoDB in one transaction with every document compared, only the changed fields sent. A run has MongoDB's own tools over MCP and POST /v1/runs/:id/mongo (find, count, insert, update_one, update_many, delete_one, delete_many), each run as SQL in its world, and SQL as before. Changes read by field path. Not yet beside MongoDB: $regex, aggregate, date ranges in filters, array elements changed one by one, masks on a field that isn't a column, sharded clusters, act_as, GATEWAY_FOLLOW. The console's connect step takes the URL.
  • MongoDB's tools, fixed before any release: an update or delete with no filter is refused (it matched every document); two updates of one document at once no longer undo each other; a sort by several fields is an array, which keeps its order; {"_id": 1} projects only _id; a string _id of 24 hex digits stays a string on insert; mongodb+srv:// URLs connect (the driver lacked DNS seed-list support).
  • Several databases behind one gateway. POST /v1/connect {name, ...} connects one more database beside the first, each with its own engine, tenant column, rules, labels and masks. A run names its database (source), and requests name theirs with X-Chronos-Source or ?source=; without one they mean the first, so nothing changes for a gateway with one database. GET /v1/sources lists them, events say source, and the console has a picker. The start-up settings, the model reviewer and Chronos Cloud's link are the first database's only.
  • High availability. With GATEWAY_STORE_URL the gateway keeps its store in a Postgres, and several gateways sharing it and one data folder take turns leading: the leader runs the engines, the others send their requests to it, and when it is lost another leads within about 15 seconds. GATEWAY_NODE_URL says how they reach each other. The Helm chart has ha.enabled. See docs/high-availability.md for what it needs and its limits. Without the setting nothing changes: SQLite, one gateway.
  • A run whose world was lost when the engine stopped hard (unmerged work isn't kept through a crash) now says so: 409 world_lost, the run failed, run.failed sent. Before, its calls answered the engine's own 404.
  • Extension points (src/ext.rs): a binary can build more into the gateway at fixed points (an approval about to merge, webhook formats, endpoints, the leader's work). The open gateway has none built in; its command line is now cli::run, which such a binary starts with its own. An approval an extension holds for more people answers waiting, and the console and the widget say so.
  • Sign-in can be built in beside OpenID Connect's (src/ext.rs: sign_in, login, person): the gateway then has sign-in without GATEWAY_OIDC_ISSUER, with who may approve and who is an admin from GATEWAY_SSO_ALLOW and GATEWAY_SSO_ADMINS (the GATEWAY_OIDC_ names still work). An extension may say who a person is to a directory, asked at each request, so access taken away there ends at once. Nothing changes for OpenID Connect.

0.2.11

With Chronos DB 0.4.103 changes

  • An AI reviewer (any OpenAI-compatible model) decides held merges before a person, seeing structure rather than row text.
  • Runs take row_rules, and the engine's new merge-policy rules.
  • The run proxy refuses SQL that switches world or merges.

0.2.10

With Chronos DB 0.4.91 change

  • Runs can write as the end user's Postgres role (act_as), so Postgres's grants and row-level security judge them.

0.2.9

With Chronos DB 0.4.81 change

  • People allowed by a group can approve; signing out asks the provider again; a refused person can try another account.

0.2.8

With Chronos DB 0.4.81 change

  • A person holding an approval link no longer sees a conflict's rows, hidden columns included.

0.2.7

With Chronos DB 0.4.61 change

  • The console, approval widget and approval pages redrawn in Signal; --chronos-on-accent.

0.2.6

With Chronos DB 0.4.61 change

  • Signed downloads and image (cosign). The installer links to the guide.

0.2.5

With Chronos DB 0.4.61 change

  • A tenant's run is held only by its own tenant's rows.

0.2.4

With Chronos DB 0.4.51 change

  • The connection check accepts providers' certificates; a person may approve a run whose reads changed.

0.2.3

With Chronos DB 0.4.51 change

  • Neon; runs that change child tables.

0.2.2

With Chronos DB 0.4.41 change

  • Fixes from a clean-machine install.

0.2.1

With Chronos DB 0.4.41 change

  • Console fixes.

0.2.0

With Chronos DB 0.4.41 change

  • The admin console.

0.1.1

With Chronos DB 0.4.20 changes

0.1.0

With Chronos DB 0.4.21 change

  • First release.
Chronos Gateway changelog · Chronos