跳到正文

Kaeso 0.7.0

140 项变更
概要

概要

Kaeso 0.7.0 adds decision models in three API formats, a rerank endpoint, guardrails, webhooks, SCIM 2.0 provisioning, key pools, batches and a Helm chart, new built-in providers from Amazon Bedrock and Azure to TypeSafe, and brings the console, documentation and UI library into the Kaeso Vanguard design. Updates from the console or the installer now check the install's settings before switching anything and stop before changing files when the new gateway would refuse one. Three settings that 0.6.0 accepted are refused at startup, so check kpmr.interval, mcp_servers.sync_interval and smtp.user before you upgrade.

  1. 01

    BreakingThe gateway stops at startup when kpmr.interval is below 5 minutes, which 0.6.0 accepted; set it to 5m or longer, or remove it for auto.

    #1120#1407
  2. 02

    BreakingThe gateway stops at startup when mcp_servers.sync_interval is below a minute, which 0.6.0 raised to a minute; set it to 1m or longer, or remove it.

    #1120#1407
  3. 03

    BreakingThe gateway stops at startup when smtp.user is set without smtp.password, where 0.6.0 sent email without authentication; add the password, or remove the user for a server without authentication.

    #1121#1407
Contributors@dvnold
新增62

新增

62 项变更Gateway, KPP, Console, UI library, Self-hosting
  1. 01

    Asking Kaeso about usage in a client that supports MCP Apps opens an interactive view with the period's totals, a chart and the top models, providers or keys, through the new show_usage_stats tool; clients without MCP Apps get a one-line summary plus the same data as structuredContent.

    #909
  2. 02

    MCP servers that ship MCP Apps views show them when a client reaches them through a project's MCP portal: the portal answers resources/list, resources/templates/list and resources/read, and reading a view needs the same right as calling a tool that names it.

    #910
  3. 03

    An identity provider can add and remove an organization's members through SCIM 2.0 at /platform/v1/orgs/{org}/scim/v2, with its own bearer token per organization created under Security › Provisioning; removing a member at the identity provider removes them as an admin would, and only accounts at the organization's verified domains are touched.

    #1244
  4. 04

    Single sign-on sets members' roles and projects from the identity provider's groups, mapped to an organization role and project memberships in Add only or Keep in sync mode, and for Microsoft Entra ID users in more than 200 groups Kaeso reads their groups from Microsoft Graph.

    #1245#1340
  5. 05

    Prices can be entered and spend read in other currencies at the ECB's daily reference rates; an organization sets the currency Kaeso shows all its money in, while everything is still stored and costed in US dollars.

    #1328#1376
  6. 06

    Projects see when their models shut down, who still calls them and what replaces them: replies from models with announced dates carry Deprecation, Sunset and Link headers, a request naming a retired model fails with 410, and aliases can move to the replacement in one click or on the shutdown day, both off by default.

    #1331
  7. 07

    A project can add model lines such as claude-opus-latest that follow the provider's newest stable model with a policy of follow (the default), ask, delay or pin; a move never happens when the project has no key for the provider or when a caller that may use the line may not call the new target.

    #1353
  8. 08

    Usage, a member's stats, a project's budgets and its summary can be shared through read-only links that people without a Kaeso account can open; each link is live or a snapshot, can hide costs and member, key and project names, expires after 1, 7, 30 or 90 days or never, and is checked against its maker's current rights each time it is opened, while request logs are never shareable; the rights usage.share and usage.share.manage govern them.

    #1347
  9. 09

    Requiring two-factor authentication first names the members it would lock out, and members hear what to do when an organization starts requiring two-factor authentication or single sign-on of them, through the new notification kind A sign-in rule affects you.

    #1355#1360
  10. 10

    Kaeso emails 7 days and 1 day before an organization's platform key expires, to members who hold platform_keys.create and to the key's creator while they are still a member.

    #1349
  11. 11

    Test key on Providers asks the provider whether it takes a key by reading its model listing, without billing a request; POST /orgs/{org}/credentials/{credential}/check answers ok, refused, unreachable or unknown.

    #1350
  12. 12

    Instance admins can list, resend and revoke invitations to the install from Users, and see when each user last signed in on the Users list and in the user sheet.

    #1342#1351
  13. 13

    Analytics breakdown tables sort by any number and download as CSV, and a project's rows open the requests they count; project and account API keys can be searched and sorted, project members searched and filtered by role, and an organization's projects sorted by their requests or cost in the last 30 days.

    #1233#1329#1341
  14. 14

    The model catalogue filters by status, an alias can be duplicated into a new one with the same route, and the route editor shows an alias's unsaved changes as a diff before saving.

    #1309#1314#1327
  15. 15

    A member's or key's limit can be copied to someone else with Copy to…, and limits sort by how much is used.

    #1317
  16. 16

    A platform key opens its details with who created it, when and the keys a rotation linked it to, and API keys can be renamed while the secret keeps working.

    #1321#1326
  17. 17

    Instance admins export the instance audit log as JSON Lines; the project audit log names the keys, models, aliases, service accounts and limits it lists and opens each, and member and user sheets open the audit log at their changes or the changes made to them.

    #1315#1332#1337
  18. 18

    The command palette opens a pasted request id in the project's request log, and the shortcut sheet lists every keyboard shortcut that works where you are, with a go-to key for every project page.

    #1320#1335
  19. 19

    Every finished Playground reply links to its request in the request log, a conversation exports as Markdown or JSON, and the last prompt can be edited and sent again to every model.

    #1322#1323#1324
  20. 20

    Expired invitations can be resent and revoked ones sent again, pending invitations can be revoked several at a time, the instance's Users list shows who has two-factor, and Organizations shows each one's sign-in requirements and filters by single sign-on.

    #1333#1334
  21. 21

    Account › Connections says when each MCP connection was last used and shows when you signed in and when it renews, and an agent's Activity names the organization and project each call acted in.

    #1336#1348

Gateway

22–49
  1. 22

    Apps make decisions through Kaeso at /v1/decisions, /v1/systemone and /v1beta/decisions on OpenAI, Google, OpenRouter, Vercel, Baseten, Fireworks and Perplexity models, with Kaeso's keys, limits, budgets, logs, costs, alias routes and response cache; a new setting body_limits.decisions (default 64 MiB) bounds their bodies, and chat, token counts and embeddings on a decision-only model are refused with a message that names the decision endpoints.

    #969
  2. 23

    TypeSafe is a built-in provider: add its key and call Jev from any decision format, with images and box questions, which are refused before sending on Jev 1.13, and Cloudflare's Clef decision models (@cf/cloudflare/clef, clef-flash and clef-omni) answer decisions in any decision format through the Cloudflare provider with the same key it uses for chat.

    #970#974
  3. 24

    Custom providers serve decisions in System One's or OpenAI's format once an org admin turns them on, with the path and, for System One, the image placement their server takes (assets, state, root or none); detection never turns decisions on, and a decision model can be added only while its decisions are on.

    #972
  4. 25

    Standard Compute and Perplexity are built-in providers: Standard Compute calls its smart routing or a pinned model from any app format, with routers listed first and marked Router in Add models, and Perplexity calls its Sonar models, with search fields such as search_mode passing for Chat Completions apps only.

    #959#966
  5. 26

    DeepInfra, Upstage, IONOS AI Model Hub, Novita AI, OVHcloud AI Endpoints, Tencent TokenHub and Venice are built-in providers: add a key and call their models and embeddings from any app format, and TokenHub's other regions work as base URL overrides.

    #1105#1107#1109#1110#1111#1117#1118
  6. 27

    Volcengine Ark, StepFun and Baidu Qianfan are built-in providers, with a new provider-level API root so Ark's /api/v3 and Qianfan's /v2 work like every other provider, Qianfan's embeddings, and StepFun's China service, Qianfan's international service and BytePlus ModelArk as base URL overrides.

    #1108#1116#1363
  7. 28

    Amazon Bedrock is a built-in provider: a Bedrock API key with its AWS region calls Claude models through the Messages route and other models through Chat Completions, requests can be signed with AWS Signature Version 4 from an access key or the gateway's own AWS role via aws.ambient_credentials (instance admins only), and models that answer only Converse, such as Amazon Nova and Meta Llama, are reached through it.

    #1174#1246#1247
  8. 29

    Azure is a built-in provider: add the key of an Azure OpenAI or Microsoft Foundry resource with its endpoint and call the resource's deployments from any app format, a key can instead be a Microsoft Entra ID service principal for resources that turn key access off, and a per-model Provider model ID sets the id Kaeso sends to the provider.

    #1187#1254
  9. 30

    Cloudflare AI Gateway is a built-in provider: a key calls one Cloudflare account through one of its AI Gateways, with Workers AI, routed models, custom models and dynamic routes, and a key without a gateway calls Workers AI directly, including its embedding models.

    #962#1106
  10. 31

    Apps rerank at /v1/rerank and /v2/rerank in Cohere's rerank format, which Cohere's Rerank 5 models answer, and Voyage AI and Jina AI are built-in providers for embeddings from any app format and rerankers through the same endpoint.

    #1211#1212#1213
  11. 32

    fal and Replicate are built-in providers for media: add a key (a token for Replicate) and make images with FLUX and Seedream and videos with Veo 3.1 and Kling from OpenAI's image and video formats, with fal also mapping Wan 3.0 and Replicate Nano Banana Pro; only the models they map by hand are accepted, and other models are refused with the list.

    #1356#1357
  12. 33

    Deepgram, Cartesia and AssemblyAI are built-in providers for speech from OpenAI's audio format: Deepgram transcribes with Nova-3 and Whisper and speaks with Aura voices, Cartesia speaks with Sonic and transcribes with Ink Whisper, and AssemblyAI transcribes with Universal-3.6 Pro through its Sync API; fields a provider has no place for are refused before sending.

    #1361#1362#1366
  13. 34

    The docs, the UI library's site and the console link a legal notice and a privacy policy set with legal.imprint_url and legal.privacy_url, each an https address; the gateway refuses to start with anything else, and an install that sets neither shows no legal links.

    #961
  14. 35

    Kaeso reads request bodies sent with Content-Encoding: gzip, x-gzip or zstd, as TypeSafe's SDKs send them, on /v1, /v1beta and /v2; other encodings such as br or deflate, and stacked ones, get 415 in the app's format, a body that decodes past the route's body limit gets 413, and a corrupt one 400.

    #963
  15. 36

    Model API replies carry the provider's request id in kaeso-upstream-request-id, non-streamed replies carry what they cost in kaeso-cost when the key's owner holds costs.view, and every answer, streams included, carries kaeso-limits, an RFC 9651 list of what is left of the key's rate limits and hard budgets, when the key's owner may see the project's limits.

    #1188#1189#1318
  16. 37

    A request can shorten how long Kaeso waits for each provider with kaeso-timeout (1 to 3600 seconds, never raising a wait), keep its content out of the request log with kaeso-log-content: off, and tag itself with up to 10 kaeso-tags that the request log filters by; browser apps may send all three, and another value answers 400.

    #1215#1218#1230
  17. 38

    cors.allowed_headers lists header names that preflights from listed origins also allow, so a browser app can send a header an alias route splits on, and server-timing is exposed to browser apps; *, cookie, forwarded, x-forwarded-… and names that aren't header names stop the gateway at startup.

    #1220
  18. 39

    The response cache honours Cache-Control: max-age, taking a stored reply only when it is at most that many seconds old, and only-if-cached, which answers from the cache or with 504 without calling the provider.

    #1308
  19. 40

    /metrics gains kaeso_update_available, the KPMR fetch and attempt timestamps including kaeso_kpmr_last_attempt_failed, kaeso_connections_open, kaeso_connections_max, kaeso_model_requests_in_flight, and kaeso_tool_calls_total with kaeso_tool_call_duration_seconds for MCP portal tool calls.

    #1207#1208#1209#1210
  20. 41

    With otel.endpoint set, each model request is exported as an OpenTelemetry trace over OTLP/HTTP JSON: a kaeso.request server span that continues the app's traceparent, a client span per route step that called a provider, and MCP portal tool calls as execute_tool spans; otel.content adds the prompt and reply where the request log keeps content, and spans never delay or fail a request.

    #1235
  21. 42

    The request log stores each long message an agent conversation repeats once per project (ClickHouse migration 0017_content_parts), so a session's stored content grows with its turns instead of their square, and every reader gets the body back byte for byte.

    #1248
  22. 43

    A route step tries its model again after a dropped connection, a 5xx or an overloaded provider, never after a timeout, a rate limit or a refused request; steps take retries from 0 to 3 and projects a Retries setting, the request log lists each try made again as retried, and kaeso-route adds tries=N.

    #1236
  23. 44

    Organizations send operational notices, such as budget alerts, limits reached, expiring keys and MCP server changes, to up to 10 webhooks signed as Standard Webhooks specifies, as Kaeso's JSON event or a Slack line; deliveries are at least once and tried again after 1, 5, 15 and 60 minutes, a receiver that fails 20 times in a row is turned off, and the rights webhooks.view and webhooks.manage govern them.

    #1237
  24. 45

    Organizations register guardrails, HTTP services of their own that Kaeso asks before a request goes to a model and, if chosen, before the reply goes back; a guardrail can allow, block or replace messages, requests are signed like Standard Webhooks, a block answers content_filtered with status 400 in the app's format, and only chat requests are checked.

    #1238
  25. 46

    Built-in guardrail rules find secrets and personal data inside Kaeso, such as provider keys, tokens, email addresses, IBANs and card numbers, plus up to 20 custom patterns, and mask them, block the request or only count them; Mask in the request log keeps the values out of the log, and Restore in the reply puts masked values back in the model's answer.

    #1240
  26. 47

    A project can spread its requests over several keys of one provider: assigning a key adds it to the project's pool with a weight from 1 to 100, a caller stays on one key, and a key gives way to the next when the failure is the key's own, such as a refused key, billing or a rate limit; video jobs, image generations, speech, realtime and ElevenLabs still use the primary key.

    #1241
  27. 48

    Responses apps can store and continue responses at OpenAI where the organization and project allow it; store: true and previous_response_id are refused everywhere else with a message that says why, conversation and background: true stay refused, and GET and DELETE /v1/responses/{response_id} work for stored responses.

    #1242
  28. 49

    Apps can run OpenAI-format batches through Kaeso at providers whose batch API takes them as they are, today OpenAI; /v1/files and /v1/batches check each line against the key and model access before anything reaches the provider, a finished batch is charged at the batch price, 50 percent for OpenAI, with one usage and request log row per line.

    #1243

KPP

50
  1. 50

    KPP reads and writes decision requests, typed questions about an input answered with probabilities, in TypeSafe's System One API, OpenAI's Decisions API and Gemini's Decisions API; requests and replies cross between all three, and what a format cannot express is refused before sending.

    #964#967#968

Console

51–58
  1. 51

    The Playground asks decision models typed questions and shows every answer with its probabilities beside chat models in the same comparison, and the Platform API gains POST …/playground/decisions and a decision mode in playground/models.

    #965#973
  2. 52

    The request log filters by alias, fallback, endpoint, minimum cost, the model or alias name the app sent (requested_model), outcome, response cache, duration and error message; GET …/logs/errors lists the most common failures, which the console shows above the list when Failed is on, and an hour under Slowest hours opens the slow requests.

    #1222#1224#1225#1231
  3. 53

    The request log offers client IP, alias, formats, response cache, fallbacks and upstream request columns, hidden at first, and the request inspector can copy a JSON request whose content and media were kept whole as a command that sends it again.

    #1226#1311
  4. 54

    The request log's cost_basis says none for requests no provider billed and estimated when Kaeso estimated the tokens; entries logged before still say reported.

    #1312
  5. 55

    The request log names each request's client app from its user agent and filters by it, the usage timeline groups by client, and both also group by alias, endpoint, status and response_cache.

    #1234#1344
  6. 56

    The log inspector links to the model, alias, key and sender it names and filters the list by each; a logged request in the Anthropic, Chat Completions, Responses or Gemini format opens in the Playground with its settings and last message, a catalogue model opens there from its sheet, and key, member, service account and model sheets open their requests and audit history.

    #1227#1228#1229
  7. 57

    The Kaeso plugin asks for the install's MCP address when it is installed in Claude Code, and for Claude on the web, desktop and mobile and for Cowork, Account › Agents has an Add to Claude button, and the plugin's skill a link, that opens Claude's custom connector form with the install's MCP address filled in.

    #1006
  8. 58

    Decision answers line their bars up under each question without repeating the answer, and the request log lists each question with what its answers mean.

    #1004

UI library

59
  1. 59

    The UI library gains ProbabilityBar, DecisionResponse and QuestionEditor.

    #976

Self-hosting

60–62
  1. 60

    Kaeso installs on Kubernetes with a Helm chart that each release publishes; it runs the gateway and console hardened as Compose does, migrates once in a Helm hook, and offers an Ingress or HTTPRoute, a PodDisruptionBudget, an optional ServiceMonitor and bundled single stores for trying it out, and the new setting updates.installed_with makes the console's Updates page show the Helm update command on such an install.

    #1232
  2. 61

    kaeso healthcheck --ready asks /readyz from inside the gateway's container, prints its answer and exits 0 only on a 200, for platforms whose readiness check runs a command; plain healthcheck stays on /healthz.

    #1205
  3. 62

    Operators can print the accounts with kaeso user list, check the settings with kaeso config check, send a test email with kaeso mail test and fetch model data with kaeso kpmr sync; Sync now on the System page does the same through POST /instance/kpmr/sync.

    #1310#1319
变更12

变更

12 项变更Gateway, KPP, UI library, Self-hosting
  1. 01

    An install takes less memory, CPU and disk: ClickHouse writes only the diagnostics an operator reads, request bodies are compressed in 16–32 MiB blocks so agent sessions are stored several times smaller, and the gateway allocates with mimalloc without transparent huge pages.

    #1217
  2. 02

    Kaeso Vanguard brings the console, docs, UI library and KPMR into the design of the redesigned kaeso.app, with cut plates, hairline structure and flat surfaces; no prop, export or variant key was removed or renamed.

    #1252
  3. 03

    Rust dependencies are updated, among them toml 1.2.0, sigstore 0.15.0, resvg 0.50.0, redis, jiff, flate2, uuid and rustls-pki-types.

    #926#927#987#988#1002#1399#1400
  4. 04

    JavaScript dependencies are updated across the console, the documentation, the UI library and the MCP view packages.

    #929#1007#1345#1398

Gateway

05–08
  1. 05

    The gateway stops at startup when kpmr.interval is below 5 minutes, when mcp_servers.sync_interval is below a minute, or when smtp.user is set without smtp.password, where 0.6.0 accepted the first two and sent email without authentication; an empty password counts as unset, and an empty KAESO_CONFIG now counts as unset instead of stopping the gateway.

    #1120#1121
  2. 06

    A request after a pause no longer waits for the database or an account renewal: caches stay warm while they are used, ChatGPT and Grok accounts are renewed ahead of time by a new account_renewal job, bodies are checked and decoded in fewer passes, and a fallback's overhead_us now leaves out the failed steps' wait for their providers.

    #1216
  3. 07

    @modelcontextprotocol/ext-apps is updated from 1.7.5 to 2.0.4 for the usage view the gateway serves.

    #930
  4. 08

    A fresh install starts with KPMR's model data as of 20 November 2026, 08:44 UTC: 2,934 models from the 50 providers Kaeso calls, with no model removed or newly deprecated.

    #1393

KPP

09
  1. 09

    Large OpenAI Responses requests, such as a coding agent's resent conversation, decode in about three quarters of the time; a 3 MB body decoded in 10.2–10.7 ms against 13.2–14.2 ms before.

    #1249

UI library

10
  1. 10

    astro is updated to 7.4.7.

    #991

Self-hosting

11–12
  1. 11

    Docker image tags are updated across the base images and deployment files, and the hashicorp/google OpenTofu provider of the GCP templates moves from 6.50.0 to 8.11.0.

    #923#985#986#1301#1302#1354#1395#1396
  2. 12

    An update from the console or the update window that changes compose.yaml, as every update from 0.6.0 does, first checks that the new gateway takes the install's settings; when it would refuse one at startup, the update stops before switching anything, the running release keeps serving, compose.yaml is put back and the update's detail names the setting, and the installer's --update does the same check.

    #1407
修复64

修复

64 项变更Gateway, KPP, Console, UI library, Documentation, Self-hosting
  1. 01

    Pages that render Kaeso UI on a server hydrate without an error when a table shows dates or relative times, a chart highlights a day, or a list starts long or with rows selected.

    #931
  2. 02

    UI library formatters follow each page's language on a server through new useFormatters() and createFormatters; StatusCode shows a focus ring, wrapped CodeBlock lines break only at spaces, and dark-theme provider marks stay legible.

    #960
  3. 03

    TypeSafe's image refusals (wrong format, too small, too large) are one sentence each, so the Playground shows their reason.

    #978
  4. 04

    A lazily loaded component asks for its code only at a fresh address after a download fails in Safari or Firefox.

    #1010
  5. 05

    Lowering the install's maximum log retention brings organizations above it down to it, and entries logged before a retention was shortened expire by the shorter retention.

    #1268#1286
  6. 06

    Emails work in more cases: a failed run of the monthly cost report or weekly summary is retried a minute later instead of skipping the month or week, with several gateways every emailed setup link works until setup is done, budgets with the period total send their alert emails, and unsubscribe links in emails sent before a master-key rotation keep working while the old key is configured.

    #1282#1283#1289#1296
  7. 07

    Reading one of your own keys needs only account.keys.view, as listing them does.

    #1278
  8. 08

    SSO group rules match Microsoft Entra ID directory roles for users whose groups are read from Microsoft Graph.

    #1343
  9. 09

    A finished video or image generation shows the tags, client app, alias, model line and paying key of the request that started it in the request log and usage, so filtering by them counts its cost.

    #1421

Gateway

10–31
  1. 10

    Gemini apps reach Kaeso under /v1 as well as /v1beta, so google-genai 3 works with its defaults, and GET /v1/models, GET /v1/models/{model} and POST /v1/decisions answer in Gemini's format when the request carries x-goog-api-key or ?key= and no Authorization header.

    #971
  2. 11

    Browser apps can send Cache-Control to the response cache; CORS preflights allow it.

    #1113
  3. 12

    Web searches that Anthropic and OpenAI bill per call are part of a request's cost from a search unit price in the catalogue; searches without a price log the cost as partly unknown.

    #1206
  4. 13

    Kaeso's own MCP server speaks protocol 2026-07-28, so newer clients connect without a failed first request: /mcp answers server/discover, resources/read carries ttlMs and cacheScope, an unknown resource answers -32602, and MCP Apps is declared as an extension.

    #1239#1250
  5. 14

    A budget counter rebuilt while a request finishes counts that request exactly once.

    #1251
  6. 15

    Costs are billed more exactly: cached tokens of a model without a cache price cost its input price instead of nothing, OpenAI's 12-hour prompt-cache writes cost their 12-hour rate, and reasoning tokens cost the catalogue's reasoning price when it lists one.

    #1253#1276#1281
  7. 16

    Thinking and sampling settings match what models accept: DeepSeek models turn thinking off when an app turns it off, Claude Opus 4.7 and later leave out top_p and top_k as they do a temperature, and a thinking budget reaches Claude models that refuse budgets as adaptive thinking, each change named in kaeso-adapted or kaeso-left-out.

    #1258#1383#1385
  8. 17

    Gemini and OpenAI Images no longer count cached image and audio tokens again as uncached input.

    #1261#1277
  9. 18

    Google's 403 for a file the request names reaches the app as Google's answer instead of a refused key, and Google's 402 that prepaid credits are depleted reaches apps as a billing refusal instead of a rate limit to retry.

    #1262#1388
  10. 19

    Gemini apps get an object in functionCall.args when another format's model writes other tool arguments, and a streamed Gemini reply whose prompt Google blocks ends with a content-filter stop instead of a server error.

    #1263#1264
  11. 20

    Request logs are written again after a ClickHouse migration instead of failing until the gateway restarts, and /readyz reports degraded while writes fail.

    #1267
  12. 21

    The MCP server no longer lists the Playground's tools, which always refused agents, the MCP usage view says 1 request for a row with one request, and agents that follow OAuth 2.1 finish signing in without sending redirect_uri to the token endpoint.

    #1269#1270#1275
  13. 22

    A model request for a provider with no key in the project answers 502 like a refused key, not 400, and a provider's 404 about a file the request names no longer holds an alias route's model for 5 minutes.

    #1272#1273
  14. 23

    Audit log, member, user and usage filters with a NUL character answer 400 naming the filter instead of 500 or 503.

    #1274
  15. 24

    A streamed reply that fails before any output falls back to the alias's next step, and a direct request answers with the error's status instead of a 200 stream with an error event, including an OpenAI Responses stream that sends a bare error event and ends.

    #1279#1378
  16. 25

    A ChatGPT account at its plan's usage limit keeps every model of the account out of routes for at least 5 minutes, or as long as the provider says, up to an hour.

    #1280
  17. 26

    The usage timeline counts a finished video or ElevenLabs image once and without a 0 ms request.

    #1288
  18. 27

    Groq json and text transcriptions are priced by Groq's own length, and audio sent as a link shows an unknown cost instead of $0.

    #1291
  19. 28

    A request body over the size limit gets Kaeso's 413 in the app's format behind a proxy such as Traefik instead of the proxy's 502, and so does an answer Kaeso sends before it has read a large body, such as for an invalid key, a wrong method or a body over the MCP limit; HTTP/2 is left alone, and a client that sends the body without waiting for 100 Continue can still get the proxy's 502.

    #1387#1409
  20. 29

    The request log shows logged JSON with its keys in the order the app sent them, so a decision's answers sit under their own questions and options.

    #1000
  21. 30

    The README names every built-in provider, and a test keeps the list in step.

    #1122
  22. 31

    kaeso instance admin-sso and kaeso sso verify-domain send the email the same change sends from the console.

    #1257

KPP

32–37
  1. 32

    Gemini no longer bills cached image and audio tokens again as uncached input.

    #994
  2. 33

    A Gemini token count with a field another format lacks is refused under the name the app sent.

    #1112
  3. 34

    OpenAI Videos' checked decoder refuses the request kinds its plain decoder refuses.

    #1123
  4. 35

    A DashScope or Gemini Interactions video job with an out-of-range time reads it as no time instead of panicking.

    #1298
  5. 36

    An image's detail reaches the other OpenAI format: Chat's image_url.detail becomes Responses' detail and back.

    #1265
  6. 37

    System One refuses a score level of null as it comes in instead of accepting a request it can't send, and OpenAI's Decisions API confidence follows TypeSafe's choice and score formulas.

    #1287

Console

38–54
  1. 38

    The request log's format filters list every format Kaeso logs, from the list they accept.

    #1221
  2. 39

    The Playground keeps models whose id starts with ~ or @, such as OpenRouter's ~…-latest and Cloudflare's @cf/… models.

    #1114
  3. 40

    Clearing a project's response cache shows its success message instead of reporting an empty answer.

    #1256
  4. 41

    The month forecast takes the last seven days' pace in the month's first week too.

    #1271
  5. 42

    Analytics' links to the request log open the UTC days Analytics counted.

    #1299
  6. 43

    Updating to a release that is no longer the newest shows the newer one on Instance › Updates instead of asking for a check the console can't do.

    #1297
  7. 44

    The key expiry picker offers no day whose start in UTC has passed, and a refused date names expires_on.

    #1259
  8. 45

    Guardrails has its own icon in the settings navigation instead of Security's.

    #1367
  9. 46

    Organization pages and settings fit their content at every width and language: the SCIM address shows once, the agent preset fits a phone, webhook URLs and MCP handles end in an ellipsis, and role descriptions get a text area.

    #1368
  10. 47

    Owners and admins who reach a project through their organization role, not as members, can choose which models their keys may call from a key's sheet, project pages name people from the organization who aren't members of the project instead of showing their user id, and a tool's sheet no longer sets its MCP server's name in a code font.

    #1413#1415#1417
  11. 48

    Project pages fit and line up at every width and language: the alias editor keeps Save on a phone, header actions wrap, tables, dialogs and sheets fit phones and tablets, values end on their column's edge, and touch targets are finger-sized.

    #1373#1386#1397
  12. 49

    Analytics fits phones and tablets: charts answer a tap, grouping is a menu on a phone, Requests stays beside the cost, and the custom period's dates show whole.

    #1401
  13. 50

    Organization, account and instance settings fit phones and tablets and line up at every width and language: Providers no longer scrolls sideways, sheet actions stay in reach, popups, filters and tables stay on screen, and switches and section actions keep their place.

    #1375#1382#1384#1402#1414
  14. 51

    Sign-in, invitations, shared links and the shell fit long addresses and names on a phone: the console no longer scrolls sideways, phone navigation gives way to dialogs, and the links under sign-in forms are big enough to tap.

    #1374#1391#1403
  15. 52

    Empty states line up with the page and take a smaller title, limits with large amounts fit their row, and the guardrail rules' sample text is a full text area again.

    #1369#1370#1371
  16. 53

    In the request logs, Tab always reaches the filters after the Models and Tools switch, and switching with the arrow keys keeps the focus on the switch.

    #1410
  17. 54

    Bar charts and chart tooltips set names such as gpt-5-mini in proportional figures, page header tabs stay whole beside a long title, and a word alone on a line of wrapped code stays whole.

    #1418#1419#1422

UI library

55–58
  1. 55

    Stat tiles that load their numbers show them at full size on phones instead of the smallest size.

    #1406
  2. 56

    Tables whose rows have a right-click menu take one Tab stop instead of one per row, and screen readers no longer meet focus stops between the rows.

    #1408
  3. 57

    On touch screens, the model and provider pickers answer a finger-sized tap around them, like every select.

    #1412
  4. 58

    Long tables and lists such as the request logs stay where you scrolled when the next page of rows loads, instead of jumping back near the top.

    #1416

Documentation

59–62
  1. 59

    The Platform API page and the tool-call route no longer say the Playground runs tools, the refusal for a caller without a person says who may call tools, and the configuration page says updates.check is read on every check and that background jobs can hold up to 11 connections outside the pool.

    #1115#1119
  2. 60

    The documentation describes what is new in 0.7.0 and what to check before upgrading to it, and its tables fit a phone.

    #1392
  3. 61

    The KPP docs record that OpenAI's Decisions API confidence follows TypeSafe's choice and score formulas.

    #1338
  4. 62

    Supported providers names the regional endpoints of QwenCloud, xAI and ElevenLabs, and the configuration page says a provider pointed at another region keeps its default region's prices.

    #1358#1359

Self-hosting

63–64
  1. 63

    With --domain :80 the installer writes KAESO_PUBLIC_URL and KAESO_SMTP_FROM for a port, from the new --public-url or, without it, a local address with the port, so the install it writes starts, and it repairs an install it wrote before.

    #1266
  2. 64

    An installer update that the new release refuses leaves compose.yaml.previous as it was, instead of a copy of the running compose.yaml.

    #1420
安全2

安全

2 项变更Gateway
  1. 01

    A key with platform access can no longer widen its own model list, Tools list or limits, or leave another key looser than itself; such changes are refused with 403 permission_denied and need a console session.

    #920

Gateway

02
  1. 02

    Email goes out through lettre 0.11.24, which fixes six SMTP advisories, RUSTSEC-2026-0560 to RUSTSEC-2026-0565, among them replies a man-in-the-middle could inject after STARTTLS.

    #1346

更多版本

所有版本

不是服务。是您自己​运行的软件。

一条命令即可在任何装有 Docker 的 Linux 服务器上安装最新版本 Kaeso 0.7.0,以及 Caddy、控制台、PostgreSQL、ClickHouse 和 Valkey。

curl -fsSLhttps://kaeso.app/install.sh | sh