github maximhq/bifrost ent-v2.6.0-base
Enterprise v2.6.0 base

48 minutes ago

What's Changed

  • dependabot fixes by @akshaydeo in #9716
  • feat: implement opencode session management with header resolution an… by @sstreichan in #7052
  • fix(mcp-logs): remove duplicate LLM request ID from log detail Request Details by @Pratham-Mishra04 in #9638
  • fix(mcp-logs): hide generic status description for ordinary executions in log details by @Pratham-Mishra04 in #9639
  • fix(mcp-logs): show virtual key once as a link in log detail request details by @Pratham-Mishra04 in #9641
  • feat(mcp-logs): show readable error cards with formatted upstream JSON in MCP log details by @Pratham-Mishra04 in #9642
  • fix(mcp): keep MCP settings input widths consistent regardless of description length by @Pratham-Mishra04 in #9643
  • fix(logstore): match MCP tool log team, customer and business-unit filters on JSON array columns by @Pratham-Mishra04 in #9644
  • feat(mcp-logs): add user, team, customer, business unit and project dimensions to MCP log filter data by @Pratham-Mishra04 in #9645
  • feat(mcp-logs): add searchable user, team, customer, business unit and project filters to the MCP logs sidebar by @Pratham-Mishra04 in #9646
  • feat(mcp-logs): add latency histogram, cancelled counts and compare_to_previous to MCP log stats by @Pratham-Mishra04 in #9647
  • feat(mcp-logs): replace stats cards with a metric strip showing sparklines and previous-period trends by @Pratham-Mishra04 in #9648
  • feat(mcp): relay session GET stream notifications through the chaosmock STDIO bridge by @Pratham-Mishra04 in #9689
  • test(mcp): add SYNC e2e harness tests for upstream tool changes across every client type by @Pratham-Mishra04 in #9690
  • test(mcp): remove redundant tool refresh e2e harness tests and narrow SET-07 to unsafe tool names by @Pratham-Mishra04 in #9691
  • fix(mcp-logs): bound embedded JSON parsing in error cards and show a message when the upstream error is empty by @Pratham-Mishra04 in #9728
  • feat(modelcatalog): datasheet entry overrides that survive upstream sync by @akshaydeo in #9349
  • feat(http): datasheet override endpoints and config.json datasheet_overrides by @akshaydeo in #9350
  • feat(ui): datasheet editor under Models by @akshaydeo in #9351
  • docs: datasheet editor guide, API reference and config schema entry by @akshaydeo in #9352
  • feat: gemini live api support by @TejasGhatte in #9353
  • feat: add support for scribe realtime events by @TejasGhatte in #9709
  • feat: goggle text to speech models by @TejasGhatte in #9694
  • fix(elevenlabs): allow Scribe transcription with wildcard virtual keys by @RAWx18 in #8746
  • fix(modelcatalog): preserve model parameters across URL sync by @Constantine3 in #8136
  • fix(configstore): make config hashes deterministic for map fields by @rojinebrahimi in #8095
  • adds token approximations using tokenizers for cancelled requests by @akshaydeo in #9238
  • cost compute for estimated tokens by @akshaydeo in #9657
  • fix: drop mcp_call items on the Responses-to-Chat fallback by @1AbdullahButt in #9733
  • [test]: make the SSRF multi-IP failure regression deterministic by @Constantine3 in #8386
  • fix: message serialization for vertex by @akshaydeo in #9807
  • fix(ui): only show the upgrading screen when the server confirms a chunk is gone by @impoiler in #9805
  • Map output format to gemini by @carlreid in #8627
  • feat(anthropic): support fast mode for Fable 5.2 on Vertex by @impoiler in #9821
  • feat(core): add Baseten provider for Model APIs and dedicated deployments by @impoiler in #9765
  • fix(modelcatalogresolver): honor SkipKeySelection before catalog-based provider resolution by @cgtarmenta in #5904
  • feat(ui): expose Baseten in the config schema and provider UI by @impoiler in #9766
  • docs: add the Baseten provider guide by @impoiler in #9767
  • fix(cli): configure coding agents and expand management API compatibility by @R-droid101 in #9158
  • feat(cli): select inference projects and support Claude Desktop headers by @R-droid101 in #9669
  • feat(cli): support Enterprise SSO sign-in over SSH and in containers by @R-droid101 in #9718
  • test(ci): run Baseten in CI and the provider harness by @impoiler in #9768
  • feat(baseten): route image generation to dedicated deployments by @impoiler in #9771
  • [fix]: Bedrock provider - serve computer tools on the Responses path through InvokeModel by @davidelct in #9840
  • feat(baseten): serve /v1/decisions through System One by @impoiler in #9772
  • feat(baseten): accept the dashboard /predict URL for image generation by @impoiler in #9814
  • feat(baseten): add rerank support for dedicated BEI deployments by @impoiler in #9816
  • [fix]: Bedrock provider - keep tool_reference blocks inside an invoke tool_result by @kunalgoesbyken in #9005
  • fix(mistral): follow the model's reasoning effort ladder by @impoiler in #9847
  • fix(logs ui): wire complexity and topics filters by @kohlivrinda in #9809
  • fix(ui): improve mobile layout of log headers and date picker by @impoiler in #9848
  • fix: upgrade A2A SDK and preserve native HTTP+JSON errors by @danpiths in #9861
  • fix: render A2A Task artifacts and history in agent logs by @danpiths in #9872
  • feat(governance): add a business unit owner column to budgets by @BearTS in #9583
  • refactor(governance): export the owned-budget reconciler for other builds by @BearTS in #9584
  • feat(ui): let a downstream build say what governs an entity's spend by @BearTS in #9585
  • fix(governance): keep a team's or customer's rate-limit window when an update sends only the maximum by @BearTS in #9770
  • fix(ui): turn a team's calendar alignment off when nothing on the sheet can align by @BearTS in #9773
  • fix(governance): start a newly added budget at zero when the operator resets usage by @BearTS in #9774
  • fix(governance): reconcile a team's budgets through the shared reconciler by @BearTS in #9775
  • fix(governance): drop a quarterly budget's quarter start when a single-budget update moves it to another window by @BearTS in #9808
  • [fix]: telemetry plugin - exclude queue wait from bifrost_upstream_latency_seconds by @sam-klaviyo in #9580
  • feat(core): regex key aliases with model id templates by @impoiler in #9862
  • feat(framework): resolve pattern key aliases in the model catalog by @impoiler in #9863
  • feat(ui): pattern rows in the key deployments table by @impoiler in #9864
  • community: add Bowmark to MCP library by @Metroxe in #9704
  • docs: pattern key aliases by @impoiler in #9865
  • test(core): add coverage for regex key alias precedence and routing by @impoiler in #9871
  • agents example by @akshaydeo in #9866
  • fix(openai): treat empty plain-text transcriptions as valid, not errors by @impoiler in #9909
  • feat(databricks): support configurable OAuth M2M scopes by @impoiler in #9820
  • feat(ui): show assigned Virtual MCPs in virtual key details sheet by @impoiler in #9907
  • fix(ui): pass triggerClassName through to multi-mode entity selector by @impoiler in #9915
  • improved agent tracing ux by @akshaydeo in #9886
  • [fix]: ElevenLabs provider - return with_timestamps character times instead of null by @abmorrill4 in #9789
  • feat(helm): add ServiceMonitor for Prometheus Operator by @charlychiu in #5281
  • feat(gemini): add native decisions support with input_file parts for audio, video, and PDFs by @Madhuvod in #9921
  • feat: add Kimchi as a first-class provider by @arnasnn in #9277
  • docs: document the Kimchi provider and add it to the config schema, CI and harness by @arnasnn in #9280
  • feat(sarvam): move chat and model listing to Token Factory v2 and add Prism decisions by @vikashmaddi-sarvam in #9600
  • feat(complexity router): add phrase suggestion pipeline package by @kohlivrinda in #9670
  • feat: add mcp_client_metadata_url to host Bifrost's client metadata document by @roroghost17 in #9898
  • fix: preserve nested custom tool name and description in chat tool conversions by @sammaji in #9869
  • docs: document mcp_client_metadata_url and add it to the Helm chart" by @roroghost17 in #9899
  • fix: mint unique call ids for gemini function calls missing ids and pair responses in order by @sammaji in #9649
  • feat: use native outputConfig structured output for supported bedrock models by @sammaji in #9951
  • fix: speech handlers fixes by @TejasGhatte in #9776
  • feat: soniox provider support by @TejasGhatte in #9829
  • bump-brotli-v1.2.8 by @akshaydeo in #10020
  • feat: realtime endpoint for soniox provider by @TejasGhatte in #9880
  • fix(plugins): download plugins only from allowlisted hosts by @akshaydeo in #10024
  • feat: gemini live api for vertex by @TejasGhatte in #9894
  • feat: ws speech stream endpoint by @TejasGhatte in #9902
  • feat: vertex live api via genai integration by @TejasGhatte in #9928
  • tests: realtime e2e tests for providers by @TejasGhatte in #9957
  • fix: test case fixes by @TejasGhatte in #10003
  • feat: elevenlabs integration, rest and passthrough by @TejasGhatte in #10011
  • feat: elevenlabs realtime sockets by @TejasGhatte in #10012
  • fix: logging for speech session requests by @TejasGhatte in #10022
  • fix(complexity router): follow-ups from the phrase suggestions review by @kohlivrinda in #10005
  • feat(model-routers): add a traffic period picker and warn about routing rules that use complexity_tier by @Madhuvod in #10021
  • feat(routing): add a Routers RBAC resource and data access control for model routers by @Madhuvod in #10025
  • fix: rewrite custom tools as function tools for all providers without custom tool support by @sammaji in #9895
  • fix: send anthropic betas as header on vertex and route bedrock tool changes to invokemodel by @sammaji in #9904
  • fix: preserve logprobs when converting between chat and responses formats by @sammaji in #9905
  • fix: strip effort-dependent unsupported sampling params for openai models on bedrock mantle by @sammaji in #9906
  • chore: updates provider harness by @sammaji in #9870
  • chore: updates provider harness and bedrock docs by @sammaji in #9952
  • fix: keep cache breakpoint on marked reminder block and drop trailing newline by @TejasGhatte in #9867
  • fix: error message in vertex provider for fetching token by @TejasGhatte in #9868
  • fix: prompt cache options parameter by @TejasGhatte in #10013
  • use tool call information for rendering messages on logs table by @akshaydeo in #10041
  • 11 15 changelog entry v2.5.2 by @akshaydeo in #10069
  • fix(openapimcp):give every tool argument a name clients accept by @kohlivrinda in #9460
  • feat(mcp):configurable openapi mcp spec size limit by @kohlivrinda in #8888
  • feat(transports): MCP builder exposing every /api operation as a tool by @kohlivrinda in #8889
  • feat(transports): serve bifrost mcp on /v1/mcp through the /api chain by @kohlivrinda in #8890
  • docs(openapi): match the spec to the handlers, add the OSS edition and an operationId contract check by @kohlivrinda in #8891
  • feat(warp): read and create through the MCP, with confirmation before writes by @kohlivrinda in #8892
  • feat(logs): GET /api/logs/{id}/trace, shared with Warp's get_request_trace by @kohlivrinda in #8925
  • feat(logs): GET /api/logs/search, semantic search shared with Warp by @kohlivrinda in #8926
  • feat(governance): GET user limits endpoints, shared with Warp's describe_user_limits by @kohlivrinda in #8927
  • feat(session): GET /api/session/me, the caller as Warp scopes it by @kohlivrinda in #8928
  • fix(bedrock): union a caller's anthropic_beta with gateway-derived betas on Converse by @fus3r in #5011
  • fix(mcp): unload the Bifrost MCP when public_base_url is cleared, redact virtual key values from Warp, and never stream tool calls by @kohlivrinda in #9944
  • test(mcp): cover the Bifrost MCP and Warp management in the MCP harness by @kohlivrinda in #9945
  • fix(auth): Basic admin credentials run as the local admin; update LIB-36 for the current sync interval rule by @kohlivrinda in #9946
  • feat(core): return the pre-hook request on websocket stream hooks by @impoiler in #9925
  • refactor(passthrough): share the header filter and the codex turn builder by @impoiler in #9926
  • [feat]: propagate x-bf-dim-* headers to child OTEL spans via trace.Attributes by @devin-williams in #3781
  • fix: re-read consent flow when auth probe sees a session the flow missed by @roroghost17 in #10074
  • feat(warp): add upgrade notice by @kohlivrinda in #10014
  • fix(governance): refuse a virtual key with two configs for one provider by @BearTS in #9901
  • fix(grant): allow an MCP client wildcard only when one of its tools survives by @BearTS in #9963
  • feat(grant): let an access intersect the levels of the caller's permits by @BearTS in #9662
  • test(grant): pass a level mode to NewAccess in existing tests by @BearTS in #9956
  • fix: speech and realtime gaps in gemini, elevenlabs, soniox by @TejasGhatte in #10071
  • fix(governance): refresh a team's cached virtual keys when its customer changes by @BearTS in #9958
  • feat(chatgpt-passthrough): relay the codex responses websocket by @impoiler in #9927
  • feat(governance): filter Codex passthrough model list to the caller's allowed models by @impoiler in #9929
  • feat(guardrails): record usage units and display-only estimated cost for external guardrail calls by @Madhuvod in #10045
  • fix: restore custom tool calls in chat completions and support custom calls across chat and responses conversions by @sammaji in #10067
  • fix: retry bedrock native structured output once with synthetic tool when schema is rejected by @sammaji in #10077
  • feat(config): add the entity access profile mode setting and its Config → Security control by @BearTS in #9664
  • chore: updates provider harness tests by @sammaji in #10068
  • fix(model router suggestions): enable non-scheduled runs via api by @kohlivrinda in #10083
  • chore(helm): sync chart with config.schema.json for 2.1.49 and add Redis metrics by @BearTS in #9881
  • fix(helm): save embedded redis-stack data to the persistent volume when auth is enabled by @BearTS in #10075
  • fix: passthrough allowlist in elevenlabs and xi-api-key by @TejasGhatte in #10093
  • chore: updates azure and huggingface provider test models by @sammaji in #10090
  • docs(routers): add Routers page and deprecate the Complexity Router by @Madhuvod in #10095
  • fix(warp): refuse model_router_ids filter in semantic searc by @kohlivrinda in #10102
  • chore(openapi): regenerate the OpenAPI bundles and drop stale docs navigation entries by @Madhuvod in #10098
  • fix(logstore): batch has_object flips on clickhouse hybrid storage by @impoiler in #10084
  • refactor(ui): move DefaultCoreConfig into its own module so e2e specs load it outside Vite by @impoiler in #10089
  • test(e2e): fix stale UI specs and tests that depend on worker state by @impoiler in #10092
  • fix(configstore): stop customer and team saves from rewriting their associations by @BearTS in #10091
  • feat(providers): add MachGen provider for image and video generation by @nishkarsh-machgen in #9625
  • test(handlers): update codex passthrough tests for new grant.NewAccess signature by @impoiler in #10105
  • fix(governance): apply the caller's row scope to model config reads by id and the unpaginated list by @BearTS in #10094
  • docs(guardrails): clarify Noma streaming behavior and update guardrail block response docs by @Madhuvod in #10104
  • chore(logstore): move the ClickHouse DeleteRelayLogs godoc onto its function by @BearTS in #10099
  • fix(security): approved KDF and password hashing in FIPS mode by @akshaydeo in #10108
  • fix(prompts): honor row-visibility scope when counting and deleting folders by @BearTS in #10100
  • fix(anthropic): forward thinking.block_binding with its beta header by @akshaydeo in #10040
  • fix(logstore): shrink retention delete batches on Postgres statement timeouts by @impoiler in #10111
  • tests: harness test fixes by @TejasGhatte in #10103
  • fix(relays): bound the Relay Logs filter values to recent rows by @BearTS in #10101
  • [feat]: Claude Code gateway sign-in (/login → Cloud gateway) by @akshaydeo in #9851
  • refactor(client-gateway): extract a client-agnostic core with Claude Code as its first profile by @akshaydeo in #9852
  • feat(client-gateway): serve client gateways under /gateway/ by @akshaydeo in #9853
  • test(datasheet): pin advisor pass billing edge cases (carries #8783 forward) by @akshaydeo in #10107
  • fix(auth): basic admin credentials no longer run as the local admin by @kohlivrinda in #10114
  • feat(client-gateway): configure gateways under oauth2_server_config.gateways by @akshaydeo in #9854
  • feat: cap the protocol version Bifrost's MCP server serves by @roroghost17 in #10088
  • feat(client-gateway): one shared device verify and a client-neutral device page by @akshaydeo in #9855
  • chore: simulate proxies and firewalls in the MCP harness by @roroghost17 in #10109
  • feat(client-gateway): forward client telemetry stamped with the Bifrost identity by @akshaydeo in #9856
  • feat(client-gateway): mark user-mode gateway requests as pre-authenticated by @BearTS in #9931
  • fix(client-gateway): run the bearer check where auth starts, inside tracing and metrics by @BearTS in #9939
  • feat(client-gateway): show the access token lifetime on the Gateways page by @BearTS in #9940
  • feat(client-gateway): refuse a gateway sign-in for a user with no access by @BearTS in #9941
  • feat(client-gateway): let the consent page deny a device sign-in by @BearTS in #9942
  • feat(client-gateway): show who can approve a sign-in on the Gateways page by @BearTS in #9943
  • feat(client-gateway): audit approving and denying a device sign-in by @BearTS in #9965
  • feat(telemetry): allow plain header values for Claude Code gateway collectors, redacted in the API by @BearTS in #10081
  • fix(client-gateway): require an admin session to enable the gateway or change its managed settings by @BearTS in #10106
  • fix(oauth): require the terminal code to approve a device sign-in from the consent page by @BearTS in #10110
  • fix(client-gateway): drain client gateway telemetry on shutdown without a config store by @BearTS in #10115
  • fix(client-gateway): harden config hash framing, telemetry endpoint validation, and Gateways page load/save states by @BearTS in #10119
  • fix: status code on unsupported models by @TejasGhatte in #10122
  • fix(bifrost mcp): deny /v1/mcp as a tool and add it to the OpenAPI spec by @kohlivrinda in #10116
  • feat(bifrost mcp): helm by @kohlivrinda in #10118
  • feat(docs):bifrost mcp by @kohlivrinda in #10121
  • docs(guardrails): clarify Noma streaming mask behavior for already-delivered text by @Madhuvod in #10123
  • tests: harness test fixes by @TejasGhatte in #10124
  • chore: updates vertex and anthropic provider tests by @sammaji in #10120
  • fix: check an alias's canonical model before injecting prompt cache markers by @bardrop in #10027
  • chore: add debug logs for MCP codex auth checks by @roroghost17 in #10117
  • fix(anthropic): stop double-counting cached prompt tokens in passthrough stream usage by @wangkanai in #6378
  • fix(bedrock): honor network_config.base_url and apply the dial policy by @akshaydeo in #10030
  • fix(governance): adopt config.json VK budgets without running budget hooks by @High-PO in #10133
  • fix: vertex model discovery for regional endpoints by @sammaji in #10127
  • feat(docs): docs + config schema - add chained profile execution, parallel input timing, secrets rule overrides, and service_tier CEL variable by @Madhuvod in #10137
  • fix: anthropic streaming with output_config.format by @akshaydeo in #10138
  • fix(oauth): use the configured access token TTL for grant expiry and edit it in seconds by @BearTS in #10135
  • fix(ui): polish datasheet editor layout and hide provider-less entries by @impoiler in #10130
  • feat(warp): render confirmation cards with the action's tool and arguments by @impoiler in #10131
  • fix: stop bifrost_active_requests leaking on streams cancelled before headers by @roroghost17 in #10134
  • fix(warp): decouple warp and bifrost mcp by @kohlivrinda in #10143
  • fix(bifrost mcp): update openapi-oss url and fix broken test by @kohlivrinda in #10145
  • docs: realtime and speech session docs by @TejasGhatte in #10129
  • [feat]: add optional stop reason column to request logs by @Constantine3 in #9890
  • fix(ui): mark access profile budgets as shared on virtual keys by @impoiler in #10148
  • fix(ui): standardize config page padding with no-padding-parent wrapper by @BearTS in #10152
  • fix: test case fixes for OTEL by @roroghost17 in #10151
  • test(warp): fix flaky test by @kohlivrinda in #10153
  • [fix]: Anthropic provider - stop double-counting cache-read tokens in streaming usage by @is911 in #5355
  • chore: fix harness cases by @TejasGhatte in #10150
  • fix(logstore): tune autovacuum on Postgres retention tables by @akshaydeo in #10136
  • fix(logstore): harden autovacuum migration tests and rollback by @impoiler in #10141
  • [fix]: Anthropic provider - keep a deferred web search pending and its late result in place on replay and streams by @akshaydeo in #10160
  • fix(transport): pass multipart bodies to transport plugins as a stream, not a copy by @akshaydeo in #9966
  • fix(files): stream Files API uploads from the request temp file by @akshaydeo in #9967
  • fix(passthrough): stream multipart uploads instead of rebuilding them in memory by @akshaydeo in #9968
  • fix(transport): stop streaming decompression from reading a released request stream by @akshaydeo in #9969
  • fix(transport): stream multipart uploads from the connection instead of pre-parsing them by @akshaydeo in #9970
  • test(harness): regression suite for streamed and buffered request bodies by @akshaydeo in #9971
  • fix(passthrough): keep the Content-Type of multipart parts sent without a filename by @akshaydeo in #10159
  • fix(modelcatalog): let allowed_models match key aliases of allowed models by @aharshit123456 in #8528
  • feat(bifrost mcp): expose /v1/mcp by @kohlivrinda in #10163
  • fix(anthropic): keep mid-conversation system reminders in one form across turns by @akshaydeo in #10162
  • fix: new flag for mid conversation system messages by @TejasGhatte in #10164
  • chore: harness test fixes by @TejasGhatte in #10174
  • fix: file data uri via genai integration by @TejasGhatte in #10158
  • fix(jobaccounting): record Seedance output tokens on the video settlement row by @impoiler in #10178
  • v2.6.0 by @akshaydeo in #10180
  • approval fix by @akshaydeo in #10183
  • fixes helm config fields tests by @akshaydeo in #10185
  • config <-> helm test fix by @akshaydeo in #10186
  • test fixes by @akshaydeo in #10198

New Contributors

Full Changelog: ent-v2.5.2-base...ent-v2.6.0-base

Don't miss a new bifrost release

NewReleases is sending notifications on new releases.