@meetkai/mka1 0.54.23 → 0.54.25
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +320 -283
- package/bin/mcp-server.js +4406 -1683
- package/bin/mcp-server.js.map +55 -19
- package/dist/commonjs/funcs/authOrgsGetById.d.ts +19 -0
- package/dist/commonjs/funcs/authOrgsGetById.d.ts.map +1 -0
- package/dist/commonjs/funcs/authOrgsGetById.js +146 -0
- package/dist/commonjs/funcs/authOrgsGetById.js.map +1 -0
- package/dist/commonjs/funcs/authOrgsUpdate.d.ts +19 -0
- package/dist/commonjs/funcs/authOrgsUpdate.d.ts.map +1 -0
- package/dist/commonjs/funcs/authOrgsUpdate.js +149 -0
- package/dist/commonjs/funcs/authOrgsUpdate.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.d.ts +19 -0
- package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.js +136 -0
- package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetRunOverview.d.ts +20 -0
- package/dist/commonjs/funcs/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetRunOverview.js +138 -0
- package/dist/commonjs/funcs/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.d.ts +19 -0
- package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.js +142 -0
- package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.d.ts +20 -0
- package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.js +142 -0
- package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsImportAgentRun.d.ts +19 -0
- package/dist/commonjs/funcs/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsImportAgentRun.js +130 -0
- package/dist/commonjs/funcs/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.d.ts +20 -0
- package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.js +146 -0
- package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/commonjs/lib/config.d.ts +2 -2
- package/dist/commonjs/lib/config.js +2 -2
- package/dist/commonjs/mcp-server/mcp-server.js +1 -1
- package/dist/commonjs/mcp-server/server.d.ts.map +1 -1
- package/dist/commonjs/mcp-server/server.js +13 -1
- package/dist/commonjs/mcp-server/server.js.map +1 -1
- package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts +8 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.js +64 -0
- package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/commonjs/models/components/appendagenttrialsrequest.d.ts +111 -0
- package/dist/commonjs/models/components/appendagenttrialsrequest.d.ts.map +1 -0
- package/dist/commonjs/models/components/appendagenttrialsrequest.js +136 -0
- package/dist/commonjs/models/components/appendagenttrialsrequest.js.map +1 -0
- package/dist/commonjs/models/components/basicorganization.d.ts +41 -0
- package/dist/commonjs/models/components/basicorganization.d.ts.map +1 -0
- package/dist/commonjs/models/components/basicorganization.js +80 -0
- package/dist/commonjs/models/components/basicorganization.js.map +1 -0
- package/dist/commonjs/models/components/evalleaderboardrow.d.ts +85 -0
- package/dist/commonjs/models/components/evalleaderboardrow.d.ts.map +1 -0
- package/dist/commonjs/models/components/evalleaderboardrow.js +129 -0
- package/dist/commonjs/models/components/evalleaderboardrow.js.map +1 -0
- package/dist/commonjs/models/components/evalrunoverview.d.ts +101 -0
- package/dist/commonjs/models/components/evalrunoverview.d.ts.map +1 -0
- package/dist/commonjs/models/components/evalrunoverview.js +150 -0
- package/dist/commonjs/models/components/evalrunoverview.js.map +1 -0
- package/dist/commonjs/models/components/evalsampleobject.d.ts +32 -0
- package/dist/commonjs/models/components/evalsampleobject.d.ts.map +1 -1
- package/dist/commonjs/models/components/evalsampleobject.js +20 -0
- package/dist/commonjs/models/components/evalsampleobject.js.map +1 -1
- package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.d.ts +32 -0
- package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.d.ts.map +1 -0
- package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.js +87 -0
- package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.js.map +1 -0
- package/dist/commonjs/models/components/evalsuiteleaderboard.d.ts +35 -0
- package/dist/commonjs/models/components/evalsuiteleaderboard.d.ts.map +1 -0
- package/dist/commonjs/models/components/evalsuiteleaderboard.js +78 -0
- package/dist/commonjs/models/components/evalsuiteleaderboard.js.map +1 -0
- package/dist/commonjs/models/components/importagentrunrequest.d.ts +164 -0
- package/dist/commonjs/models/components/importagentrunrequest.d.ts.map +1 -0
- package/dist/commonjs/models/components/importagentrunrequest.js +182 -0
- package/dist/commonjs/models/components/importagentrunrequest.js.map +1 -0
- package/dist/commonjs/models/components/index.d.ts +14 -0
- package/dist/commonjs/models/components/index.d.ts.map +1 -1
- package/dist/commonjs/models/components/index.js +14 -0
- package/dist/commonjs/models/components/index.js.map +1 -1
- package/dist/commonjs/models/components/organizationcodeconfig.d.ts +20 -0
- package/dist/commonjs/models/components/organizationcodeconfig.d.ts.map +1 -0
- package/dist/commonjs/models/components/organizationcodeconfig.js +58 -0
- package/dist/commonjs/models/components/organizationcodeconfig.js.map +1 -0
- package/dist/commonjs/models/components/organizationmetadata.d.ts +23 -0
- package/dist/commonjs/models/components/organizationmetadata.d.ts.map +1 -0
- package/dist/commonjs/models/components/organizationmetadata.js +69 -0
- package/dist/commonjs/models/components/organizationmetadata.js.map +1 -0
- package/dist/commonjs/models/components/organizationmetadataoutput.d.ts +23 -0
- package/dist/commonjs/models/components/organizationmetadataoutput.d.ts.map +1 -0
- package/dist/commonjs/models/components/organizationmetadataoutput.js +69 -0
- package/dist/commonjs/models/components/organizationmetadataoutput.js.map +1 -0
- package/dist/commonjs/models/components/schema0.d.ts +17 -0
- package/dist/commonjs/models/components/schema0.d.ts.map +1 -0
- package/dist/commonjs/models/components/schema0.js +66 -0
- package/dist/commonjs/models/components/schema0.js.map +1 -0
- package/dist/commonjs/models/components/schema1.d.ts +17 -0
- package/dist/commonjs/models/components/schema1.d.ts.map +1 -0
- package/dist/commonjs/models/components/schema1.js +66 -0
- package/dist/commonjs/models/components/schema1.js.map +1 -0
- package/dist/commonjs/models/components/updateorganizationbody.d.ts +18 -0
- package/dist/commonjs/models/components/updateorganizationbody.d.ts.map +1 -0
- package/dist/commonjs/models/components/updateorganizationbody.js +59 -0
- package/dist/commonjs/models/components/updateorganizationbody.js.map +1 -0
- package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.d.ts +34 -0
- package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.d.ts.map +1 -0
- package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.js +71 -0
- package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.js.map +1 -0
- package/dist/commonjs/models/errors/getjwtfromkey.d.ts +9 -9
- package/dist/commonjs/models/errors/getjwtfromkey.d.ts.map +1 -1
- package/dist/commonjs/models/errors/getjwtfromkey.js +11 -11
- package/dist/commonjs/models/errors/getjwtfromkey.js.map +1 -1
- package/dist/commonjs/models/errors/getorgbyid.d.ts +211 -0
- package/dist/commonjs/models/errors/getorgbyid.d.ts.map +1 -0
- package/dist/commonjs/models/errors/getorgbyid.js +298 -0
- package/dist/commonjs/models/errors/getorgbyid.js.map +1 -0
- package/dist/commonjs/models/errors/index.d.ts +2 -0
- package/dist/commonjs/models/errors/index.d.ts.map +1 -1
- package/dist/commonjs/models/errors/index.js +2 -0
- package/dist/commonjs/models/errors/index.js.map +1 -1
- package/dist/commonjs/models/errors/updateorganization.d.ts +257 -0
- package/dist/commonjs/models/errors/updateorganization.d.ts.map +1 -0
- package/dist/commonjs/models/errors/updateorganization.js +353 -0
- package/dist/commonjs/models/errors/updateorganization.js.map +1 -0
- package/dist/commonjs/models/operations/appendagentevaltrials.d.ts +264 -0
- package/dist/commonjs/models/operations/appendagentevaltrials.d.ts.map +1 -0
- package/dist/commonjs/models/operations/appendagentevaltrials.js +371 -0
- package/dist/commonjs/models/operations/appendagentevaltrials.js.map +1 -0
- package/dist/commonjs/models/operations/cancelevalrun.d.ts +20 -0
- package/dist/commonjs/models/operations/cancelevalrun.d.ts.map +1 -1
- package/dist/commonjs/models/operations/cancelevalrun.js +16 -0
- package/dist/commonjs/models/operations/cancelevalrun.js.map +1 -1
- package/dist/commonjs/models/operations/createevalrun.d.ts +20 -0
- package/dist/commonjs/models/operations/createevalrun.d.ts.map +1 -1
- package/dist/commonjs/models/operations/createevalrun.js +16 -0
- package/dist/commonjs/models/operations/createevalrun.js.map +1 -1
- package/dist/commonjs/models/operations/getevalrun.d.ts +20 -0
- package/dist/commonjs/models/operations/getevalrun.d.ts.map +1 -1
- package/dist/commonjs/models/operations/getevalrun.js +16 -0
- package/dist/commonjs/models/operations/getevalrun.js.map +1 -1
- package/dist/commonjs/models/operations/getevalrunoverview.d.ts +24 -0
- package/dist/commonjs/models/operations/getevalrunoverview.d.ts.map +1 -0
- package/dist/commonjs/models/operations/getevalrunoverview.js +73 -0
- package/dist/commonjs/models/operations/getevalrunoverview.js.map +1 -0
- package/dist/commonjs/models/operations/getevalsampletrajectory.d.ts +75 -0
- package/dist/commonjs/models/operations/getevalsampletrajectory.d.ts.map +1 -0
- package/dist/commonjs/models/operations/getevalsampletrajectory.js +133 -0
- package/dist/commonjs/models/operations/getevalsampletrajectory.js.map +1 -0
- package/dist/commonjs/models/operations/getevalsuiteleaderboard.d.ts +32 -0
- package/dist/commonjs/models/operations/getevalsuiteleaderboard.d.ts.map +1 -0
- package/dist/commonjs/models/operations/getevalsuiteleaderboard.js +83 -0
- package/dist/commonjs/models/operations/getevalsuiteleaderboard.js.map +1 -0
- package/dist/commonjs/models/operations/getorgbyid.d.ts +37 -0
- package/dist/commonjs/models/operations/getorgbyid.d.ts.map +1 -0
- package/dist/commonjs/models/operations/getorgbyid.js +87 -0
- package/dist/commonjs/models/operations/getorgbyid.js.map +1 -0
- package/dist/commonjs/models/operations/importagentevalrun.d.ts +262 -0
- package/dist/commonjs/models/operations/importagentevalrun.d.ts.map +1 -0
- package/dist/commonjs/models/operations/importagentevalrun.js +367 -0
- package/dist/commonjs/models/operations/importagentevalrun.js.map +1 -0
- package/dist/commonjs/models/operations/index.d.ts +8 -0
- package/dist/commonjs/models/operations/index.d.ts.map +1 -1
- package/dist/commonjs/models/operations/index.js +8 -0
- package/dist/commonjs/models/operations/index.js.map +1 -1
- package/dist/commonjs/models/operations/listevalruns.d.ts +20 -0
- package/dist/commonjs/models/operations/listevalruns.d.ts.map +1 -1
- package/dist/commonjs/models/operations/listevalruns.js +16 -0
- package/dist/commonjs/models/operations/listevalruns.js.map +1 -1
- package/dist/commonjs/models/operations/listevalscheduleruns.d.ts +20 -0
- package/dist/commonjs/models/operations/listevalscheduleruns.d.ts.map +1 -1
- package/dist/commonjs/models/operations/listevalscheduleruns.js +16 -0
- package/dist/commonjs/models/operations/listevalscheduleruns.js.map +1 -1
- package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.d.ts +20 -0
- package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.d.ts.map +1 -1
- package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.js +16 -0
- package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.js.map +1 -1
- package/dist/commonjs/models/operations/rerunfailedevalsamples.d.ts +20 -0
- package/dist/commonjs/models/operations/rerunfailedevalsamples.d.ts.map +1 -1
- package/dist/commonjs/models/operations/rerunfailedevalsamples.js +16 -0
- package/dist/commonjs/models/operations/rerunfailedevalsamples.js.map +1 -1
- package/dist/commonjs/models/operations/retryfailedevalrun.d.ts +20 -0
- package/dist/commonjs/models/operations/retryfailedevalrun.d.ts.map +1 -1
- package/dist/commonjs/models/operations/retryfailedevalrun.js +16 -0
- package/dist/commonjs/models/operations/retryfailedevalrun.js.map +1 -1
- package/dist/commonjs/models/operations/updateorganization.d.ts +40 -0
- package/dist/commonjs/models/operations/updateorganization.d.ts.map +1 -0
- package/dist/commonjs/models/operations/updateorganization.js +92 -0
- package/dist/commonjs/models/operations/updateorganization.js.map +1 -0
- package/dist/commonjs/models/operations/uploadevalsampletrajectory.d.ts +31 -0
- package/dist/commonjs/models/operations/uploadevalsampletrajectory.d.ts.map +1 -0
- package/dist/commonjs/models/operations/uploadevalsampletrajectory.js +86 -0
- package/dist/commonjs/models/operations/uploadevalsampletrajectory.js.map +1 -0
- package/dist/commonjs/react-query/authOrgsGetById.core.d.ts +15 -0
- package/dist/commonjs/react-query/authOrgsGetById.core.d.ts.map +1 -0
- package/dist/commonjs/react-query/authOrgsGetById.core.js +36 -0
- package/dist/commonjs/react-query/authOrgsGetById.core.js.map +1 -0
- package/dist/commonjs/react-query/authOrgsGetById.d.ts +39 -0
- package/dist/commonjs/react-query/authOrgsGetById.d.ts.map +1 -0
- package/dist/commonjs/react-query/authOrgsGetById.js +60 -0
- package/dist/commonjs/react-query/authOrgsGetById.js.map +1 -0
- package/dist/commonjs/react-query/authOrgsUpdate.d.ts +31 -0
- package/dist/commonjs/react-query/authOrgsUpdate.d.ts.map +1 -0
- package/dist/commonjs/react-query/authOrgsUpdate.js +47 -0
- package/dist/commonjs/react-query/authOrgsUpdate.js.map +1 -0
- package/dist/commonjs/react-query/index.d.ts +8 -0
- package/dist/commonjs/react-query/index.d.ts.map +1 -1
- package/dist/commonjs/react-query/index.js +8 -0
- package/dist/commonjs/react-query/index.js.map +1 -1
- package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.d.ts +28 -0
- package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.js +47 -0
- package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.d.ts +16 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.js +37 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.d.ts +40 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.js +60 -0
- package/dist/commonjs/react-query/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.d.ts +16 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.js +45 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.d.ts +44 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.js +65 -0
- package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts +20 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.js +41 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.d.ts +48 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.js +65 -0
- package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsImportAgentRun.d.ts +28 -0
- package/dist/commonjs/react-query/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsImportAgentRun.js +47 -0
- package/dist/commonjs/react-query/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.d.ts +29 -0
- package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.js +47 -0
- package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/commonjs/sdk/evals.d.ts +42 -0
- package/dist/commonjs/sdk/evals.d.ts.map +1 -1
- package/dist/commonjs/sdk/evals.js +60 -0
- package/dist/commonjs/sdk/evals.js.map +1 -1
- package/dist/commonjs/sdk/orgs.d.ts +15 -0
- package/dist/commonjs/sdk/orgs.d.ts.map +1 -1
- package/dist/commonjs/sdk/orgs.js +20 -0
- package/dist/commonjs/sdk/orgs.js.map +1 -1
- package/dist/esm/funcs/authOrgsGetById.d.ts +19 -0
- package/dist/esm/funcs/authOrgsGetById.d.ts.map +1 -0
- package/dist/esm/funcs/authOrgsGetById.js +110 -0
- package/dist/esm/funcs/authOrgsGetById.js.map +1 -0
- package/dist/esm/funcs/authOrgsUpdate.d.ts +19 -0
- package/dist/esm/funcs/authOrgsUpdate.d.ts.map +1 -0
- package/dist/esm/funcs/authOrgsUpdate.js +113 -0
- package/dist/esm/funcs/authOrgsUpdate.js.map +1 -0
- package/dist/esm/funcs/llmEvalsAppendAgentTrials.d.ts +19 -0
- package/dist/esm/funcs/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsAppendAgentTrials.js +100 -0
- package/dist/esm/funcs/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/esm/funcs/llmEvalsGetRunOverview.d.ts +20 -0
- package/dist/esm/funcs/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsGetRunOverview.js +102 -0
- package/dist/esm/funcs/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/esm/funcs/llmEvalsGetSampleTrajectory.d.ts +19 -0
- package/dist/esm/funcs/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsGetSampleTrajectory.js +106 -0
- package/dist/esm/funcs/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.d.ts +20 -0
- package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.js +106 -0
- package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/esm/funcs/llmEvalsImportAgentRun.d.ts +19 -0
- package/dist/esm/funcs/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsImportAgentRun.js +94 -0
- package/dist/esm/funcs/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.d.ts +20 -0
- package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.js +110 -0
- package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/esm/lib/config.d.ts +2 -2
- package/dist/esm/lib/config.js +2 -2
- package/dist/esm/mcp-server/mcp-server.js +1 -1
- package/dist/esm/mcp-server/server.d.ts.map +1 -1
- package/dist/esm/mcp-server/server.js +13 -1
- package/dist/esm/mcp-server/server.js.map +1 -1
- package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts +8 -0
- package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.js +28 -0
- package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/esm/models/components/appendagenttrialsrequest.d.ts +111 -0
- package/dist/esm/models/components/appendagenttrialsrequest.d.ts.map +1 -0
- package/dist/esm/models/components/appendagenttrialsrequest.js +96 -0
- package/dist/esm/models/components/appendagenttrialsrequest.js.map +1 -0
- package/dist/esm/models/components/basicorganization.d.ts +41 -0
- package/dist/esm/models/components/basicorganization.d.ts.map +1 -0
- package/dist/esm/models/components/basicorganization.js +42 -0
- package/dist/esm/models/components/basicorganization.js.map +1 -0
- package/dist/esm/models/components/evalleaderboardrow.d.ts +85 -0
- package/dist/esm/models/components/evalleaderboardrow.d.ts.map +1 -0
- package/dist/esm/models/components/evalleaderboardrow.js +89 -0
- package/dist/esm/models/components/evalleaderboardrow.js.map +1 -0
- package/dist/esm/models/components/evalrunoverview.d.ts +101 -0
- package/dist/esm/models/components/evalrunoverview.d.ts.map +1 -0
- package/dist/esm/models/components/evalrunoverview.js +108 -0
- package/dist/esm/models/components/evalrunoverview.js.map +1 -0
- package/dist/esm/models/components/evalsampleobject.d.ts +32 -0
- package/dist/esm/models/components/evalsampleobject.d.ts.map +1 -1
- package/dist/esm/models/components/evalsampleobject.js +20 -0
- package/dist/esm/models/components/evalsampleobject.js.map +1 -1
- package/dist/esm/models/components/evalsampletrajectoryuploadedobject.d.ts +32 -0
- package/dist/esm/models/components/evalsampletrajectoryuploadedobject.d.ts.map +1 -0
- package/dist/esm/models/components/evalsampletrajectoryuploadedobject.js +49 -0
- package/dist/esm/models/components/evalsampletrajectoryuploadedobject.js.map +1 -0
- package/dist/esm/models/components/evalsuiteleaderboard.d.ts +35 -0
- package/dist/esm/models/components/evalsuiteleaderboard.d.ts.map +1 -0
- package/dist/esm/models/components/evalsuiteleaderboard.js +40 -0
- package/dist/esm/models/components/evalsuiteleaderboard.js.map +1 -0
- package/dist/esm/models/components/importagentrunrequest.d.ts +164 -0
- package/dist/esm/models/components/importagentrunrequest.d.ts.map +1 -0
- package/dist/esm/models/components/importagentrunrequest.js +142 -0
- package/dist/esm/models/components/importagentrunrequest.js.map +1 -0
- package/dist/esm/models/components/index.d.ts +14 -0
- package/dist/esm/models/components/index.d.ts.map +1 -1
- package/dist/esm/models/components/index.js +14 -0
- package/dist/esm/models/components/index.js.map +1 -1
- package/dist/esm/models/components/organizationcodeconfig.d.ts +20 -0
- package/dist/esm/models/components/organizationcodeconfig.d.ts.map +1 -0
- package/dist/esm/models/components/organizationcodeconfig.js +20 -0
- package/dist/esm/models/components/organizationcodeconfig.js.map +1 -0
- package/dist/esm/models/components/organizationmetadata.d.ts +23 -0
- package/dist/esm/models/components/organizationmetadata.d.ts.map +1 -0
- package/dist/esm/models/components/organizationmetadata.js +31 -0
- package/dist/esm/models/components/organizationmetadata.js.map +1 -0
- package/dist/esm/models/components/organizationmetadataoutput.d.ts +23 -0
- package/dist/esm/models/components/organizationmetadataoutput.d.ts.map +1 -0
- package/dist/esm/models/components/organizationmetadataoutput.js +31 -0
- package/dist/esm/models/components/organizationmetadataoutput.js.map +1 -0
- package/dist/esm/models/components/schema0.d.ts +17 -0
- package/dist/esm/models/components/schema0.d.ts.map +1 -0
- package/dist/esm/models/components/schema0.js +28 -0
- package/dist/esm/models/components/schema0.js.map +1 -0
- package/dist/esm/models/components/schema1.d.ts +17 -0
- package/dist/esm/models/components/schema1.d.ts.map +1 -0
- package/dist/esm/models/components/schema1.js +28 -0
- package/dist/esm/models/components/schema1.js.map +1 -0
- package/dist/esm/models/components/updateorganizationbody.d.ts +18 -0
- package/dist/esm/models/components/updateorganizationbody.d.ts.map +1 -0
- package/dist/esm/models/components/updateorganizationbody.js +21 -0
- package/dist/esm/models/components/updateorganizationbody.js.map +1 -0
- package/dist/esm/models/components/uploadevalsampletrajectoryrequest.d.ts +34 -0
- package/dist/esm/models/components/uploadevalsampletrajectoryrequest.d.ts.map +1 -0
- package/dist/esm/models/components/uploadevalsampletrajectoryrequest.js +33 -0
- package/dist/esm/models/components/uploadevalsampletrajectoryrequest.js.map +1 -0
- package/dist/esm/models/errors/getjwtfromkey.d.ts +9 -9
- package/dist/esm/models/errors/getjwtfromkey.d.ts.map +1 -1
- package/dist/esm/models/errors/getjwtfromkey.js +8 -8
- package/dist/esm/models/errors/getjwtfromkey.js.map +1 -1
- package/dist/esm/models/errors/getorgbyid.d.ts +211 -0
- package/dist/esm/models/errors/getorgbyid.d.ts.map +1 -0
- package/dist/esm/models/errors/getorgbyid.js +254 -0
- package/dist/esm/models/errors/getorgbyid.js.map +1 -0
- package/dist/esm/models/errors/index.d.ts +2 -0
- package/dist/esm/models/errors/index.d.ts.map +1 -1
- package/dist/esm/models/errors/index.js +2 -0
- package/dist/esm/models/errors/index.js.map +1 -1
- package/dist/esm/models/errors/updateorganization.d.ts +257 -0
- package/dist/esm/models/errors/updateorganization.d.ts.map +1 -0
- package/dist/esm/models/errors/updateorganization.js +306 -0
- package/dist/esm/models/errors/updateorganization.js.map +1 -0
- package/dist/esm/models/operations/appendagentevaltrials.d.ts +264 -0
- package/dist/esm/models/operations/appendagentevaltrials.d.ts.map +1 -0
- package/dist/esm/models/operations/appendagentevaltrials.js +323 -0
- package/dist/esm/models/operations/appendagentevaltrials.js.map +1 -0
- package/dist/esm/models/operations/cancelevalrun.d.ts +20 -0
- package/dist/esm/models/operations/cancelevalrun.d.ts.map +1 -1
- package/dist/esm/models/operations/cancelevalrun.js +16 -0
- package/dist/esm/models/operations/cancelevalrun.js.map +1 -1
- package/dist/esm/models/operations/createevalrun.d.ts +20 -0
- package/dist/esm/models/operations/createevalrun.d.ts.map +1 -1
- package/dist/esm/models/operations/createevalrun.js +16 -0
- package/dist/esm/models/operations/createevalrun.js.map +1 -1
- package/dist/esm/models/operations/getevalrun.d.ts +20 -0
- package/dist/esm/models/operations/getevalrun.d.ts.map +1 -1
- package/dist/esm/models/operations/getevalrun.js +16 -0
- package/dist/esm/models/operations/getevalrun.js.map +1 -1
- package/dist/esm/models/operations/getevalrunoverview.d.ts +24 -0
- package/dist/esm/models/operations/getevalrunoverview.d.ts.map +1 -0
- package/dist/esm/models/operations/getevalrunoverview.js +35 -0
- package/dist/esm/models/operations/getevalrunoverview.js.map +1 -0
- package/dist/esm/models/operations/getevalsampletrajectory.d.ts +75 -0
- package/dist/esm/models/operations/getevalsampletrajectory.d.ts.map +1 -0
- package/dist/esm/models/operations/getevalsampletrajectory.js +93 -0
- package/dist/esm/models/operations/getevalsampletrajectory.js.map +1 -0
- package/dist/esm/models/operations/getevalsuiteleaderboard.d.ts +32 -0
- package/dist/esm/models/operations/getevalsuiteleaderboard.d.ts.map +1 -0
- package/dist/esm/models/operations/getevalsuiteleaderboard.js +45 -0
- package/dist/esm/models/operations/getevalsuiteleaderboard.js.map +1 -0
- package/dist/esm/models/operations/getorgbyid.d.ts +37 -0
- package/dist/esm/models/operations/getorgbyid.d.ts.map +1 -0
- package/dist/esm/models/operations/getorgbyid.js +47 -0
- package/dist/esm/models/operations/getorgbyid.js.map +1 -0
- package/dist/esm/models/operations/importagentevalrun.d.ts +262 -0
- package/dist/esm/models/operations/importagentevalrun.d.ts.map +1 -0
- package/dist/esm/models/operations/importagentevalrun.js +319 -0
- package/dist/esm/models/operations/importagentevalrun.js.map +1 -0
- package/dist/esm/models/operations/index.d.ts +8 -0
- package/dist/esm/models/operations/index.d.ts.map +1 -1
- package/dist/esm/models/operations/index.js +8 -0
- package/dist/esm/models/operations/index.js.map +1 -1
- package/dist/esm/models/operations/listevalruns.d.ts +20 -0
- package/dist/esm/models/operations/listevalruns.d.ts.map +1 -1
- package/dist/esm/models/operations/listevalruns.js +16 -0
- package/dist/esm/models/operations/listevalruns.js.map +1 -1
- package/dist/esm/models/operations/listevalscheduleruns.d.ts +20 -0
- package/dist/esm/models/operations/listevalscheduleruns.d.ts.map +1 -1
- package/dist/esm/models/operations/listevalscheduleruns.js +16 -0
- package/dist/esm/models/operations/listevalscheduleruns.js.map +1 -1
- package/dist/esm/models/operations/listevalsuiteleaderboardruns.d.ts +20 -0
- package/dist/esm/models/operations/listevalsuiteleaderboardruns.d.ts.map +1 -1
- package/dist/esm/models/operations/listevalsuiteleaderboardruns.js +16 -0
- package/dist/esm/models/operations/listevalsuiteleaderboardruns.js.map +1 -1
- package/dist/esm/models/operations/rerunfailedevalsamples.d.ts +20 -0
- package/dist/esm/models/operations/rerunfailedevalsamples.d.ts.map +1 -1
- package/dist/esm/models/operations/rerunfailedevalsamples.js +16 -0
- package/dist/esm/models/operations/rerunfailedevalsamples.js.map +1 -1
- package/dist/esm/models/operations/retryfailedevalrun.d.ts +20 -0
- package/dist/esm/models/operations/retryfailedevalrun.d.ts.map +1 -1
- package/dist/esm/models/operations/retryfailedevalrun.js +16 -0
- package/dist/esm/models/operations/retryfailedevalrun.js.map +1 -1
- package/dist/esm/models/operations/updateorganization.d.ts +40 -0
- package/dist/esm/models/operations/updateorganization.d.ts.map +1 -0
- package/dist/esm/models/operations/updateorganization.js +52 -0
- package/dist/esm/models/operations/updateorganization.js.map +1 -0
- package/dist/esm/models/operations/uploadevalsampletrajectory.d.ts +31 -0
- package/dist/esm/models/operations/uploadevalsampletrajectory.d.ts.map +1 -0
- package/dist/esm/models/operations/uploadevalsampletrajectory.js +48 -0
- package/dist/esm/models/operations/uploadevalsampletrajectory.js.map +1 -0
- package/dist/esm/react-query/authOrgsGetById.core.d.ts +15 -0
- package/dist/esm/react-query/authOrgsGetById.core.d.ts.map +1 -0
- package/dist/esm/react-query/authOrgsGetById.core.js +31 -0
- package/dist/esm/react-query/authOrgsGetById.core.js.map +1 -0
- package/dist/esm/react-query/authOrgsGetById.d.ts +39 -0
- package/dist/esm/react-query/authOrgsGetById.d.ts.map +1 -0
- package/dist/esm/react-query/authOrgsGetById.js +50 -0
- package/dist/esm/react-query/authOrgsGetById.js.map +1 -0
- package/dist/esm/react-query/authOrgsUpdate.d.ts +31 -0
- package/dist/esm/react-query/authOrgsUpdate.d.ts.map +1 -0
- package/dist/esm/react-query/authOrgsUpdate.js +42 -0
- package/dist/esm/react-query/authOrgsUpdate.js.map +1 -0
- package/dist/esm/react-query/index.d.ts +8 -0
- package/dist/esm/react-query/index.d.ts.map +1 -1
- package/dist/esm/react-query/index.js +8 -0
- package/dist/esm/react-query/index.js.map +1 -1
- package/dist/esm/react-query/llmEvalsAppendAgentTrials.d.ts +28 -0
- package/dist/esm/react-query/llmEvalsAppendAgentTrials.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsAppendAgentTrials.js +42 -0
- package/dist/esm/react-query/llmEvalsAppendAgentTrials.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.core.d.ts +16 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.core.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.core.js +32 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.core.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.d.ts +40 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.js +50 -0
- package/dist/esm/react-query/llmEvalsGetRunOverview.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.d.ts +16 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.js +40 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.d.ts +44 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.js +55 -0
- package/dist/esm/react-query/llmEvalsGetSampleTrajectory.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts +20 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.js +36 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.js.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.d.ts +48 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.js +55 -0
- package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.js.map +1 -0
- package/dist/esm/react-query/llmEvalsImportAgentRun.d.ts +28 -0
- package/dist/esm/react-query/llmEvalsImportAgentRun.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsImportAgentRun.js +42 -0
- package/dist/esm/react-query/llmEvalsImportAgentRun.js.map +1 -0
- package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.d.ts +29 -0
- package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
- package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.js +42 -0
- package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.js.map +1 -0
- package/dist/esm/sdk/evals.d.ts +42 -0
- package/dist/esm/sdk/evals.d.ts.map +1 -1
- package/dist/esm/sdk/evals.js +60 -0
- package/dist/esm/sdk/evals.js.map +1 -1
- package/dist/esm/sdk/orgs.d.ts +15 -0
- package/dist/esm/sdk/orgs.d.ts.map +1 -1
- package/dist/esm/sdk/orgs.js +20 -0
- package/dist/esm/sdk/orgs.js.map +1 -1
- package/jsr.json +1 -1
- package/package.json +1 -1
- package/src/funcs/authOrgsGetById.ts +251 -0
- package/src/funcs/authOrgsUpdate.ts +264 -0
- package/src/funcs/llmEvalsAppendAgentTrials.ts +188 -0
- package/src/funcs/llmEvalsGetRunOverview.ts +192 -0
- package/src/funcs/llmEvalsGetSampleTrajectory.ts +197 -0
- package/src/funcs/llmEvalsGetSuiteLeaderboard.ts +197 -0
- package/src/funcs/llmEvalsImportAgentRun.ts +181 -0
- package/src/funcs/llmEvalsUploadSampleTrajectory.ts +201 -0
- package/src/lib/config.ts +2 -2
- package/src/mcp-server/mcp-server.ts +1 -1
- package/src/mcp-server/server.ts +13 -1
- package/src/mcp-server/tools/llmEvalsAppendAgentTrials.ts +37 -0
- package/src/mcp-server/tools/llmEvalsGetRunOverview.ts +37 -0
- package/src/mcp-server/tools/llmEvalsGetSampleTrajectory.ts +37 -0
- package/src/mcp-server/tools/llmEvalsGetSuiteLeaderboard.ts +37 -0
- package/src/mcp-server/tools/llmEvalsImportAgentRun.ts +37 -0
- package/src/mcp-server/tools/llmEvalsUploadSampleTrajectory.ts +38 -0
- package/src/models/components/appendagenttrialsrequest.ts +224 -0
- package/src/models/components/basicorganization.ts +100 -0
- package/src/models/components/evalleaderboardrow.ts +200 -0
- package/src/models/components/evalrunoverview.ts +240 -0
- package/src/models/components/evalsampleobject.ts +48 -0
- package/src/models/components/evalsampletrajectoryuploadedobject.ts +98 -0
- package/src/models/components/evalsuiteleaderboard.ts +93 -0
- package/src/models/components/importagentrunrequest.ts +319 -0
- package/src/models/components/index.ts +14 -0
- package/src/models/components/organizationcodeconfig.ts +54 -0
- package/src/models/components/organizationmetadata.ts +80 -0
- package/src/models/components/organizationmetadataoutput.ts +80 -0
- package/src/models/components/schema0.ts +55 -0
- package/src/models/components/schema1.ts +55 -0
- package/src/models/components/updateorganizationbody.ts +57 -0
- package/src/models/components/uploadevalsampletrajectoryrequest.ts +79 -0
- package/src/models/errors/getjwtfromkey.ts +28 -21
- package/src/models/errors/getorgbyid.ts +499 -0
- package/src/models/errors/index.ts +2 -0
- package/src/models/errors/updateorganization.ts +631 -0
- package/src/models/operations/appendagentevaltrials.ts +678 -0
- package/src/models/operations/cancelevalrun.ts +36 -0
- package/src/models/operations/createevalrun.ts +36 -0
- package/src/models/operations/getevalrun.ts +36 -0
- package/src/models/operations/getevalrunoverview.ts +73 -0
- package/src/models/operations/getevalsampletrajectory.ts +199 -0
- package/src/models/operations/getevalsuiteleaderboard.ts +93 -0
- package/src/models/operations/getorgbyid.ts +114 -0
- package/src/models/operations/importagentevalrun.ts +665 -0
- package/src/models/operations/index.ts +8 -0
- package/src/models/operations/listevalruns.ts +36 -0
- package/src/models/operations/listevalscheduleruns.ts +36 -0
- package/src/models/operations/listevalsuiteleaderboardruns.ts +36 -0
- package/src/models/operations/rerunfailedevalsamples.ts +36 -0
- package/src/models/operations/retryfailedevalrun.ts +36 -0
- package/src/models/operations/updateorganization.ts +121 -0
- package/src/models/operations/uploadevalsampletrajectory.ts +98 -0
- package/src/react-query/authOrgsGetById.core.ts +78 -0
- package/src/react-query/authOrgsGetById.ts +148 -0
- package/src/react-query/authOrgsUpdate.ts +122 -0
- package/src/react-query/index.ts +8 -0
- package/src/react-query/llmEvalsAppendAgentTrials.ts +109 -0
- package/src/react-query/llmEvalsGetRunOverview.core.ts +77 -0
- package/src/react-query/llmEvalsGetRunOverview.ts +150 -0
- package/src/react-query/llmEvalsGetSampleTrajectory.core.ts +94 -0
- package/src/react-query/llmEvalsGetSampleTrajectory.ts +159 -0
- package/src/react-query/llmEvalsGetSuiteLeaderboard.core.ts +89 -0
- package/src/react-query/llmEvalsGetSuiteLeaderboard.ts +163 -0
- package/src/react-query/llmEvalsImportAgentRun.ts +109 -0
- package/src/react-query/llmEvalsUploadSampleTrajectory.ts +110 -0
- package/src/sdk/evals.ts +108 -0
- package/src/sdk/orgs.ts +41 -0
|
@@ -0,0 +1,224 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
|
|
5
|
+
import * as z from "zod/v3";
|
|
6
|
+
import { remap as remap$ } from "../../lib/primitives.js";
|
|
7
|
+
import { safeParse } from "../../lib/schemas.js";
|
|
8
|
+
import { ClosedEnum } from "../../types/enums.js";
|
|
9
|
+
import { Result as SafeParseResult } from "../../types/fp.js";
|
|
10
|
+
import { SDKValidationError } from "../errors/sdkvalidationerror.js";
|
|
11
|
+
|
|
12
|
+
/**
|
|
13
|
+
* `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
|
|
14
|
+
*/
|
|
15
|
+
export const AppendAgentTrialsRequestStatus = {
|
|
16
|
+
Completed: "completed",
|
|
17
|
+
Failed: "failed",
|
|
18
|
+
} as const;
|
|
19
|
+
/**
|
|
20
|
+
* `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
|
|
21
|
+
*/
|
|
22
|
+
export type AppendAgentTrialsRequestStatus = ClosedEnum<
|
|
23
|
+
typeof AppendAgentTrialsRequestStatus
|
|
24
|
+
>;
|
|
25
|
+
|
|
26
|
+
export type AppendAgentTrialsRequestTrials = {
|
|
27
|
+
/**
|
|
28
|
+
* Must already exist in the target suite version's manifest; unknown ids are rejected rather than invented.
|
|
29
|
+
*/
|
|
30
|
+
taskId: string;
|
|
31
|
+
sampleIndex: number;
|
|
32
|
+
/**
|
|
33
|
+
* `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
|
|
34
|
+
*/
|
|
35
|
+
status?: AppendAgentTrialsRequestStatus | undefined;
|
|
36
|
+
/**
|
|
37
|
+
* The task instruction.
|
|
38
|
+
*/
|
|
39
|
+
prompt?: string | undefined;
|
|
40
|
+
target?: string | null | undefined;
|
|
41
|
+
/**
|
|
42
|
+
* What decided the score — the verifier's output. The agent's session goes to the trajectory route.
|
|
43
|
+
*/
|
|
44
|
+
outputText?: string | null | undefined;
|
|
45
|
+
/**
|
|
46
|
+
* Metric values for this trial, e.g. `{ "reward": 1, "resolved": 1 }`. The board averages `reward` by default.
|
|
47
|
+
*/
|
|
48
|
+
scores?: { [k: string]: number } | undefined;
|
|
49
|
+
/**
|
|
50
|
+
* The exception this trial recorded, if any. Does not by itself make the trial a failure.
|
|
51
|
+
*/
|
|
52
|
+
error?: { [k: string]: any } | null | undefined;
|
|
53
|
+
costUsd?: number | null | undefined;
|
|
54
|
+
inputTokens?: number | null | undefined;
|
|
55
|
+
outputTokens?: number | null | undefined;
|
|
56
|
+
/**
|
|
57
|
+
* The share of `input_tokens` served from the provider's prompt cache.
|
|
58
|
+
*/
|
|
59
|
+
cachedTokens?: number | null | undefined;
|
|
60
|
+
phaseTimings?: { [k: string]: any } | null | undefined;
|
|
61
|
+
};
|
|
62
|
+
|
|
63
|
+
export type AppendAgentTrialsRequest = {
|
|
64
|
+
/**
|
|
65
|
+
* Each trial pins its own `sample_index` (unique per task and model): re-sending a batch replaces the same rows instead of duplicating them, so a retried batch is safe.
|
|
66
|
+
*/
|
|
67
|
+
trials?: Array<AppendAgentTrialsRequestTrials> | undefined;
|
|
68
|
+
/**
|
|
69
|
+
* Marks the run completed once these trials land — the last call of a streaming job sends this, usually with an empty `trials`.
|
|
70
|
+
*/
|
|
71
|
+
finalize?: boolean | undefined;
|
|
72
|
+
};
|
|
73
|
+
|
|
74
|
+
/** @internal */
|
|
75
|
+
export const AppendAgentTrialsRequestStatus$inboundSchema: z.ZodNativeEnum<
|
|
76
|
+
typeof AppendAgentTrialsRequestStatus
|
|
77
|
+
> = z.nativeEnum(AppendAgentTrialsRequestStatus);
|
|
78
|
+
/** @internal */
|
|
79
|
+
export const AppendAgentTrialsRequestStatus$outboundSchema: z.ZodNativeEnum<
|
|
80
|
+
typeof AppendAgentTrialsRequestStatus
|
|
81
|
+
> = AppendAgentTrialsRequestStatus$inboundSchema;
|
|
82
|
+
|
|
83
|
+
/** @internal */
|
|
84
|
+
export const AppendAgentTrialsRequestTrials$inboundSchema: z.ZodType<
|
|
85
|
+
AppendAgentTrialsRequestTrials,
|
|
86
|
+
z.ZodTypeDef,
|
|
87
|
+
unknown
|
|
88
|
+
> = z.object({
|
|
89
|
+
task_id: z.string(),
|
|
90
|
+
sample_index: z.number().int(),
|
|
91
|
+
status: AppendAgentTrialsRequestStatus$inboundSchema.default("completed"),
|
|
92
|
+
prompt: z.string().default(""),
|
|
93
|
+
target: z.nullable(z.string()).optional(),
|
|
94
|
+
output_text: z.nullable(z.string()).optional(),
|
|
95
|
+
scores: z.record(z.number()).optional(),
|
|
96
|
+
error: z.nullable(z.record(z.any())).optional(),
|
|
97
|
+
cost_usd: z.nullable(z.number()).optional(),
|
|
98
|
+
input_tokens: z.nullable(z.number().int()).optional(),
|
|
99
|
+
output_tokens: z.nullable(z.number().int()).optional(),
|
|
100
|
+
cached_tokens: z.nullable(z.number().int()).optional(),
|
|
101
|
+
phase_timings: z.nullable(z.record(z.any())).optional(),
|
|
102
|
+
}).transform((v) => {
|
|
103
|
+
return remap$(v, {
|
|
104
|
+
"task_id": "taskId",
|
|
105
|
+
"sample_index": "sampleIndex",
|
|
106
|
+
"output_text": "outputText",
|
|
107
|
+
"cost_usd": "costUsd",
|
|
108
|
+
"input_tokens": "inputTokens",
|
|
109
|
+
"output_tokens": "outputTokens",
|
|
110
|
+
"cached_tokens": "cachedTokens",
|
|
111
|
+
"phase_timings": "phaseTimings",
|
|
112
|
+
});
|
|
113
|
+
});
|
|
114
|
+
/** @internal */
|
|
115
|
+
export type AppendAgentTrialsRequestTrials$Outbound = {
|
|
116
|
+
task_id: string;
|
|
117
|
+
sample_index: number;
|
|
118
|
+
status: string;
|
|
119
|
+
prompt: string;
|
|
120
|
+
target?: string | null | undefined;
|
|
121
|
+
output_text?: string | null | undefined;
|
|
122
|
+
scores?: { [k: string]: number } | undefined;
|
|
123
|
+
error?: { [k: string]: any } | null | undefined;
|
|
124
|
+
cost_usd?: number | null | undefined;
|
|
125
|
+
input_tokens?: number | null | undefined;
|
|
126
|
+
output_tokens?: number | null | undefined;
|
|
127
|
+
cached_tokens?: number | null | undefined;
|
|
128
|
+
phase_timings?: { [k: string]: any } | null | undefined;
|
|
129
|
+
};
|
|
130
|
+
|
|
131
|
+
/** @internal */
|
|
132
|
+
export const AppendAgentTrialsRequestTrials$outboundSchema: z.ZodType<
|
|
133
|
+
AppendAgentTrialsRequestTrials$Outbound,
|
|
134
|
+
z.ZodTypeDef,
|
|
135
|
+
AppendAgentTrialsRequestTrials
|
|
136
|
+
> = z.object({
|
|
137
|
+
taskId: z.string(),
|
|
138
|
+
sampleIndex: z.number().int(),
|
|
139
|
+
status: AppendAgentTrialsRequestStatus$outboundSchema.default("completed"),
|
|
140
|
+
prompt: z.string().default(""),
|
|
141
|
+
target: z.nullable(z.string()).optional(),
|
|
142
|
+
outputText: z.nullable(z.string()).optional(),
|
|
143
|
+
scores: z.record(z.number()).optional(),
|
|
144
|
+
error: z.nullable(z.record(z.any())).optional(),
|
|
145
|
+
costUsd: z.nullable(z.number()).optional(),
|
|
146
|
+
inputTokens: z.nullable(z.number().int()).optional(),
|
|
147
|
+
outputTokens: z.nullable(z.number().int()).optional(),
|
|
148
|
+
cachedTokens: z.nullable(z.number().int()).optional(),
|
|
149
|
+
phaseTimings: z.nullable(z.record(z.any())).optional(),
|
|
150
|
+
}).transform((v) => {
|
|
151
|
+
return remap$(v, {
|
|
152
|
+
taskId: "task_id",
|
|
153
|
+
sampleIndex: "sample_index",
|
|
154
|
+
outputText: "output_text",
|
|
155
|
+
costUsd: "cost_usd",
|
|
156
|
+
inputTokens: "input_tokens",
|
|
157
|
+
outputTokens: "output_tokens",
|
|
158
|
+
cachedTokens: "cached_tokens",
|
|
159
|
+
phaseTimings: "phase_timings",
|
|
160
|
+
});
|
|
161
|
+
});
|
|
162
|
+
|
|
163
|
+
export function appendAgentTrialsRequestTrialsToJSON(
|
|
164
|
+
appendAgentTrialsRequestTrials: AppendAgentTrialsRequestTrials,
|
|
165
|
+
): string {
|
|
166
|
+
return JSON.stringify(
|
|
167
|
+
AppendAgentTrialsRequestTrials$outboundSchema.parse(
|
|
168
|
+
appendAgentTrialsRequestTrials,
|
|
169
|
+
),
|
|
170
|
+
);
|
|
171
|
+
}
|
|
172
|
+
export function appendAgentTrialsRequestTrialsFromJSON(
|
|
173
|
+
jsonString: string,
|
|
174
|
+
): SafeParseResult<AppendAgentTrialsRequestTrials, SDKValidationError> {
|
|
175
|
+
return safeParse(
|
|
176
|
+
jsonString,
|
|
177
|
+
(x) => AppendAgentTrialsRequestTrials$inboundSchema.parse(JSON.parse(x)),
|
|
178
|
+
`Failed to parse 'AppendAgentTrialsRequestTrials' from JSON`,
|
|
179
|
+
);
|
|
180
|
+
}
|
|
181
|
+
|
|
182
|
+
/** @internal */
|
|
183
|
+
export const AppendAgentTrialsRequest$inboundSchema: z.ZodType<
|
|
184
|
+
AppendAgentTrialsRequest,
|
|
185
|
+
z.ZodTypeDef,
|
|
186
|
+
unknown
|
|
187
|
+
> = z.object({
|
|
188
|
+
trials: z.array(z.lazy(() => AppendAgentTrialsRequestTrials$inboundSchema))
|
|
189
|
+
.optional(),
|
|
190
|
+
finalize: z.boolean().default(false),
|
|
191
|
+
});
|
|
192
|
+
/** @internal */
|
|
193
|
+
export type AppendAgentTrialsRequest$Outbound = {
|
|
194
|
+
trials?: Array<AppendAgentTrialsRequestTrials$Outbound> | undefined;
|
|
195
|
+
finalize: boolean;
|
|
196
|
+
};
|
|
197
|
+
|
|
198
|
+
/** @internal */
|
|
199
|
+
export const AppendAgentTrialsRequest$outboundSchema: z.ZodType<
|
|
200
|
+
AppendAgentTrialsRequest$Outbound,
|
|
201
|
+
z.ZodTypeDef,
|
|
202
|
+
AppendAgentTrialsRequest
|
|
203
|
+
> = z.object({
|
|
204
|
+
trials: z.array(z.lazy(() => AppendAgentTrialsRequestTrials$outboundSchema))
|
|
205
|
+
.optional(),
|
|
206
|
+
finalize: z.boolean().default(false),
|
|
207
|
+
});
|
|
208
|
+
|
|
209
|
+
export function appendAgentTrialsRequestToJSON(
|
|
210
|
+
appendAgentTrialsRequest: AppendAgentTrialsRequest,
|
|
211
|
+
): string {
|
|
212
|
+
return JSON.stringify(
|
|
213
|
+
AppendAgentTrialsRequest$outboundSchema.parse(appendAgentTrialsRequest),
|
|
214
|
+
);
|
|
215
|
+
}
|
|
216
|
+
export function appendAgentTrialsRequestFromJSON(
|
|
217
|
+
jsonString: string,
|
|
218
|
+
): SafeParseResult<AppendAgentTrialsRequest, SDKValidationError> {
|
|
219
|
+
return safeParse(
|
|
220
|
+
jsonString,
|
|
221
|
+
(x) => AppendAgentTrialsRequest$inboundSchema.parse(JSON.parse(x)),
|
|
222
|
+
`Failed to parse 'AppendAgentTrialsRequest' from JSON`,
|
|
223
|
+
);
|
|
224
|
+
}
|
|
@@ -0,0 +1,100 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
|
|
5
|
+
import * as z from "zod/v3";
|
|
6
|
+
import { safeParse } from "../../lib/schemas.js";
|
|
7
|
+
import { ClosedEnum } from "../../types/enums.js";
|
|
8
|
+
import { Result as SafeParseResult } from "../../types/fp.js";
|
|
9
|
+
import { SDKValidationError } from "../errors/sdkvalidationerror.js";
|
|
10
|
+
import {
|
|
11
|
+
OrganizationMetadataOutput,
|
|
12
|
+
OrganizationMetadataOutput$inboundSchema,
|
|
13
|
+
OrganizationMetadataOutput$Outbound,
|
|
14
|
+
OrganizationMetadataOutput$outboundSchema,
|
|
15
|
+
} from "./organizationmetadataoutput.js";
|
|
16
|
+
|
|
17
|
+
export const BasicOrganizationStatus = {
|
|
18
|
+
Active: "active",
|
|
19
|
+
Suspended: "suspended",
|
|
20
|
+
Deleted: "deleted",
|
|
21
|
+
} as const;
|
|
22
|
+
export type BasicOrganizationStatus = ClosedEnum<
|
|
23
|
+
typeof BasicOrganizationStatus
|
|
24
|
+
>;
|
|
25
|
+
|
|
26
|
+
export type BasicOrganization = {
|
|
27
|
+
id: string;
|
|
28
|
+
name: string;
|
|
29
|
+
slug: string;
|
|
30
|
+
logo: string | null;
|
|
31
|
+
createdAt: string;
|
|
32
|
+
status: BasicOrganizationStatus;
|
|
33
|
+
metadata: OrganizationMetadataOutput | null;
|
|
34
|
+
};
|
|
35
|
+
|
|
36
|
+
/** @internal */
|
|
37
|
+
export const BasicOrganizationStatus$inboundSchema: z.ZodNativeEnum<
|
|
38
|
+
typeof BasicOrganizationStatus
|
|
39
|
+
> = z.nativeEnum(BasicOrganizationStatus);
|
|
40
|
+
/** @internal */
|
|
41
|
+
export const BasicOrganizationStatus$outboundSchema: z.ZodNativeEnum<
|
|
42
|
+
typeof BasicOrganizationStatus
|
|
43
|
+
> = BasicOrganizationStatus$inboundSchema;
|
|
44
|
+
|
|
45
|
+
/** @internal */
|
|
46
|
+
export const BasicOrganization$inboundSchema: z.ZodType<
|
|
47
|
+
BasicOrganization,
|
|
48
|
+
z.ZodTypeDef,
|
|
49
|
+
unknown
|
|
50
|
+
> = z.object({
|
|
51
|
+
id: z.string(),
|
|
52
|
+
name: z.string(),
|
|
53
|
+
slug: z.string(),
|
|
54
|
+
logo: z.nullable(z.string()),
|
|
55
|
+
createdAt: z.string(),
|
|
56
|
+
status: BasicOrganizationStatus$inboundSchema,
|
|
57
|
+
metadata: z.nullable(OrganizationMetadataOutput$inboundSchema),
|
|
58
|
+
});
|
|
59
|
+
/** @internal */
|
|
60
|
+
export type BasicOrganization$Outbound = {
|
|
61
|
+
id: string;
|
|
62
|
+
name: string;
|
|
63
|
+
slug: string;
|
|
64
|
+
logo: string | null;
|
|
65
|
+
createdAt: string;
|
|
66
|
+
status: string;
|
|
67
|
+
metadata: OrganizationMetadataOutput$Outbound | null;
|
|
68
|
+
};
|
|
69
|
+
|
|
70
|
+
/** @internal */
|
|
71
|
+
export const BasicOrganization$outboundSchema: z.ZodType<
|
|
72
|
+
BasicOrganization$Outbound,
|
|
73
|
+
z.ZodTypeDef,
|
|
74
|
+
BasicOrganization
|
|
75
|
+
> = z.object({
|
|
76
|
+
id: z.string(),
|
|
77
|
+
name: z.string(),
|
|
78
|
+
slug: z.string(),
|
|
79
|
+
logo: z.nullable(z.string()),
|
|
80
|
+
createdAt: z.string(),
|
|
81
|
+
status: BasicOrganizationStatus$outboundSchema,
|
|
82
|
+
metadata: z.nullable(OrganizationMetadataOutput$outboundSchema),
|
|
83
|
+
});
|
|
84
|
+
|
|
85
|
+
export function basicOrganizationToJSON(
|
|
86
|
+
basicOrganization: BasicOrganization,
|
|
87
|
+
): string {
|
|
88
|
+
return JSON.stringify(
|
|
89
|
+
BasicOrganization$outboundSchema.parse(basicOrganization),
|
|
90
|
+
);
|
|
91
|
+
}
|
|
92
|
+
export function basicOrganizationFromJSON(
|
|
93
|
+
jsonString: string,
|
|
94
|
+
): SafeParseResult<BasicOrganization, SDKValidationError> {
|
|
95
|
+
return safeParse(
|
|
96
|
+
jsonString,
|
|
97
|
+
(x) => BasicOrganization$inboundSchema.parse(JSON.parse(x)),
|
|
98
|
+
`Failed to parse 'BasicOrganization' from JSON`,
|
|
99
|
+
);
|
|
100
|
+
}
|
|
@@ -0,0 +1,200 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
|
|
5
|
+
import * as z from "zod/v3";
|
|
6
|
+
import { remap as remap$ } from "../../lib/primitives.js";
|
|
7
|
+
import { safeParse } from "../../lib/schemas.js";
|
|
8
|
+
import { Result as SafeParseResult } from "../../types/fp.js";
|
|
9
|
+
import { SDKValidationError } from "../errors/sdkvalidationerror.js";
|
|
10
|
+
|
|
11
|
+
export type AccuracyCi = {
|
|
12
|
+
low: number;
|
|
13
|
+
high: number;
|
|
14
|
+
margin: number;
|
|
15
|
+
};
|
|
16
|
+
|
|
17
|
+
export type EvalLeaderboardRow = {
|
|
18
|
+
agentName: string;
|
|
19
|
+
/**
|
|
20
|
+
* Version from the row's newest run. Not part of the row identity, so older runs in this row may have reported a different one.
|
|
21
|
+
*/
|
|
22
|
+
agentVersion: string | null;
|
|
23
|
+
agentEffort: string | null;
|
|
24
|
+
model: string;
|
|
25
|
+
/**
|
|
26
|
+
* Mean of `metric` over the row's scored trials. Null when no trial carried a numeric value for that metric — distinct from 0, which means every trial scored zero.
|
|
27
|
+
*/
|
|
28
|
+
accuracy: number | null;
|
|
29
|
+
/**
|
|
30
|
+
* 95% Wald interval, p ± 1.96·sqrt(p(1-p)/n), clamped to [0, 1]. Derived per request rather than stored, so it cannot drift when a failed trial is rerun. It assumes `metric` is a 0/1 outcome (the harness `reward`); on a continuous metric the mean is still right but this interval is not, so read it only for pass/fail metrics. Null when `accuracy` is.
|
|
31
|
+
*/
|
|
32
|
+
accuracyCi: AccuracyCi | null;
|
|
33
|
+
/**
|
|
34
|
+
* Every sample in the row's runs, whatever its status.
|
|
35
|
+
*/
|
|
36
|
+
trials: number;
|
|
37
|
+
/**
|
|
38
|
+
* Trials carrying a numeric `metric` — the denominator behind `accuracy` and `accuracy_ci`.
|
|
39
|
+
*/
|
|
40
|
+
scoredTrials: number;
|
|
41
|
+
/**
|
|
42
|
+
* Trials that ended in the `failed` status, i.e. produced no score at all. An agent timeout is NOT one of these: the harness still verifies and scores that trial.
|
|
43
|
+
*/
|
|
44
|
+
failedTrials: number;
|
|
45
|
+
/**
|
|
46
|
+
* Summed trial spend, falling back to the runs' own totals when trials carry no cost. Null when neither reported any.
|
|
47
|
+
*/
|
|
48
|
+
costUsd: number | null;
|
|
49
|
+
runCount: number;
|
|
50
|
+
/**
|
|
51
|
+
* The runs behind this row, largest (most trials) first, ties newest first — element 0 is the run to open when the reader drills into this row, so a full benchmark job outranks a newer one-task smoke run.
|
|
52
|
+
*/
|
|
53
|
+
runIds: Array<string>;
|
|
54
|
+
/**
|
|
55
|
+
* Creation time of the newest run in the row, unix seconds.
|
|
56
|
+
*/
|
|
57
|
+
lastRunAt: number;
|
|
58
|
+
};
|
|
59
|
+
|
|
60
|
+
/** @internal */
|
|
61
|
+
export const AccuracyCi$inboundSchema: z.ZodType<
|
|
62
|
+
AccuracyCi,
|
|
63
|
+
z.ZodTypeDef,
|
|
64
|
+
unknown
|
|
65
|
+
> = z.object({
|
|
66
|
+
low: z.number(),
|
|
67
|
+
high: z.number(),
|
|
68
|
+
margin: z.number(),
|
|
69
|
+
});
|
|
70
|
+
/** @internal */
|
|
71
|
+
export type AccuracyCi$Outbound = {
|
|
72
|
+
low: number;
|
|
73
|
+
high: number;
|
|
74
|
+
margin: number;
|
|
75
|
+
};
|
|
76
|
+
|
|
77
|
+
/** @internal */
|
|
78
|
+
export const AccuracyCi$outboundSchema: z.ZodType<
|
|
79
|
+
AccuracyCi$Outbound,
|
|
80
|
+
z.ZodTypeDef,
|
|
81
|
+
AccuracyCi
|
|
82
|
+
> = z.object({
|
|
83
|
+
low: z.number(),
|
|
84
|
+
high: z.number(),
|
|
85
|
+
margin: z.number(),
|
|
86
|
+
});
|
|
87
|
+
|
|
88
|
+
export function accuracyCiToJSON(accuracyCi: AccuracyCi): string {
|
|
89
|
+
return JSON.stringify(AccuracyCi$outboundSchema.parse(accuracyCi));
|
|
90
|
+
}
|
|
91
|
+
export function accuracyCiFromJSON(
|
|
92
|
+
jsonString: string,
|
|
93
|
+
): SafeParseResult<AccuracyCi, SDKValidationError> {
|
|
94
|
+
return safeParse(
|
|
95
|
+
jsonString,
|
|
96
|
+
(x) => AccuracyCi$inboundSchema.parse(JSON.parse(x)),
|
|
97
|
+
`Failed to parse 'AccuracyCi' from JSON`,
|
|
98
|
+
);
|
|
99
|
+
}
|
|
100
|
+
|
|
101
|
+
/** @internal */
|
|
102
|
+
export const EvalLeaderboardRow$inboundSchema: z.ZodType<
|
|
103
|
+
EvalLeaderboardRow,
|
|
104
|
+
z.ZodTypeDef,
|
|
105
|
+
unknown
|
|
106
|
+
> = z.object({
|
|
107
|
+
agent_name: z.string(),
|
|
108
|
+
agent_version: z.nullable(z.string()),
|
|
109
|
+
agent_effort: z.nullable(z.string()),
|
|
110
|
+
model: z.string(),
|
|
111
|
+
accuracy: z.nullable(z.number()),
|
|
112
|
+
accuracy_ci: z.nullable(z.lazy(() => AccuracyCi$inboundSchema)),
|
|
113
|
+
trials: z.number().int(),
|
|
114
|
+
scored_trials: z.number().int(),
|
|
115
|
+
failed_trials: z.number().int(),
|
|
116
|
+
cost_usd: z.nullable(z.number()),
|
|
117
|
+
run_count: z.number().int(),
|
|
118
|
+
run_ids: z.array(z.string()),
|
|
119
|
+
last_run_at: z.number().int(),
|
|
120
|
+
}).transform((v) => {
|
|
121
|
+
return remap$(v, {
|
|
122
|
+
"agent_name": "agentName",
|
|
123
|
+
"agent_version": "agentVersion",
|
|
124
|
+
"agent_effort": "agentEffort",
|
|
125
|
+
"accuracy_ci": "accuracyCi",
|
|
126
|
+
"scored_trials": "scoredTrials",
|
|
127
|
+
"failed_trials": "failedTrials",
|
|
128
|
+
"cost_usd": "costUsd",
|
|
129
|
+
"run_count": "runCount",
|
|
130
|
+
"run_ids": "runIds",
|
|
131
|
+
"last_run_at": "lastRunAt",
|
|
132
|
+
});
|
|
133
|
+
});
|
|
134
|
+
/** @internal */
|
|
135
|
+
export type EvalLeaderboardRow$Outbound = {
|
|
136
|
+
agent_name: string;
|
|
137
|
+
agent_version: string | null;
|
|
138
|
+
agent_effort: string | null;
|
|
139
|
+
model: string;
|
|
140
|
+
accuracy: number | null;
|
|
141
|
+
accuracy_ci: AccuracyCi$Outbound | null;
|
|
142
|
+
trials: number;
|
|
143
|
+
scored_trials: number;
|
|
144
|
+
failed_trials: number;
|
|
145
|
+
cost_usd: number | null;
|
|
146
|
+
run_count: number;
|
|
147
|
+
run_ids: Array<string>;
|
|
148
|
+
last_run_at: number;
|
|
149
|
+
};
|
|
150
|
+
|
|
151
|
+
/** @internal */
|
|
152
|
+
export const EvalLeaderboardRow$outboundSchema: z.ZodType<
|
|
153
|
+
EvalLeaderboardRow$Outbound,
|
|
154
|
+
z.ZodTypeDef,
|
|
155
|
+
EvalLeaderboardRow
|
|
156
|
+
> = z.object({
|
|
157
|
+
agentName: z.string(),
|
|
158
|
+
agentVersion: z.nullable(z.string()),
|
|
159
|
+
agentEffort: z.nullable(z.string()),
|
|
160
|
+
model: z.string(),
|
|
161
|
+
accuracy: z.nullable(z.number()),
|
|
162
|
+
accuracyCi: z.nullable(z.lazy(() => AccuracyCi$outboundSchema)),
|
|
163
|
+
trials: z.number().int(),
|
|
164
|
+
scoredTrials: z.number().int(),
|
|
165
|
+
failedTrials: z.number().int(),
|
|
166
|
+
costUsd: z.nullable(z.number()),
|
|
167
|
+
runCount: z.number().int(),
|
|
168
|
+
runIds: z.array(z.string()),
|
|
169
|
+
lastRunAt: z.number().int(),
|
|
170
|
+
}).transform((v) => {
|
|
171
|
+
return remap$(v, {
|
|
172
|
+
agentName: "agent_name",
|
|
173
|
+
agentVersion: "agent_version",
|
|
174
|
+
agentEffort: "agent_effort",
|
|
175
|
+
accuracyCi: "accuracy_ci",
|
|
176
|
+
scoredTrials: "scored_trials",
|
|
177
|
+
failedTrials: "failed_trials",
|
|
178
|
+
costUsd: "cost_usd",
|
|
179
|
+
runCount: "run_count",
|
|
180
|
+
runIds: "run_ids",
|
|
181
|
+
lastRunAt: "last_run_at",
|
|
182
|
+
});
|
|
183
|
+
});
|
|
184
|
+
|
|
185
|
+
export function evalLeaderboardRowToJSON(
|
|
186
|
+
evalLeaderboardRow: EvalLeaderboardRow,
|
|
187
|
+
): string {
|
|
188
|
+
return JSON.stringify(
|
|
189
|
+
EvalLeaderboardRow$outboundSchema.parse(evalLeaderboardRow),
|
|
190
|
+
);
|
|
191
|
+
}
|
|
192
|
+
export function evalLeaderboardRowFromJSON(
|
|
193
|
+
jsonString: string,
|
|
194
|
+
): SafeParseResult<EvalLeaderboardRow, SDKValidationError> {
|
|
195
|
+
return safeParse(
|
|
196
|
+
jsonString,
|
|
197
|
+
(x) => EvalLeaderboardRow$inboundSchema.parse(JSON.parse(x)),
|
|
198
|
+
`Failed to parse 'EvalLeaderboardRow' from JSON`,
|
|
199
|
+
);
|
|
200
|
+
}
|