What's Changed
- Remove grading from User Testing by @Paoli99 in #5122
- Generate cases breadcrumb by @ignaciojimenezr in #5125
- Eval UI updates: deletion, generation UX, compare deltas by @mintlify[bot] in #5134
- Fix server groups hidden when named after their only server by @chelojimenez in #5135
- Remove the retired failure-groups card by @chelojimenez in #5132
- fix(desktop): retire the Update button when the download never lands by @ZeHuari in #4946
- BB-234: move the Swarms server warning onto the page, in informational blue by @SebasKoria in #5124
- swarms: disable promote on sessions whose run did not succeed (BB-247) by @SebasKoria in #5127
- Publish the evals action's tags from the release workflow by @ignaciojimenezr in #5140
- Send the service token with the forwarded guest IP hash by @prathmeshpatel in #5145
- Playground: Enable BYOK multi-model columns by @sophiemarinavo in #5146
- Put the live swarm matrix on the run URL by @chelojimenez in #5151
- Swarm run detail: Run tab + header outcome chip by @mintlify[bot] in #5152
- Fix the red audio-auth tests on main by @ignaciojimenezr in #5154
- Fix the SDK README's first test assertion by @ignaciojimenezr in #5144
- Document executeTool return type and narrowing helpers by @mintlify[bot] in #5158
- fix: prompt guests to sign up before sharing by @prathmeshpatel in #5155
- Raise the credits wall from the Markdown import, and give BYOK a destination by @ignaciojimenezr in #5157
- Update BYOK out-of-credits button behavior docs by @mintlify[bot] in #5160
- Surface a refused guest mint as a 429 and offer sign-in by @chelojimenez in #5156
- Document guest session rate limit banner by @mintlify[bot] in #5161
- Link eval PR reports at the Ding Dong tab by @ignaciojimenezr in #5149
- fix(evals): find a suite's server when a host supplies it by @ignaciojimenezr in #5147
- Classify the OAuth consent state, and restyle the error card by @chelojimenez in #5141
- Inspector: bill journeys by run kind and show plan limits by @prathmeshpatel in #5162
- Remove redundant Analyze findings button by @ignaciojimenezr in #5163
- Recognize the backend's free_tier_model_restricted denial by @chelojimenez in #5142
- ops(alerts): version the 2026-09-16 spend, fan-out and E2B monitors by @chelojimenez in #5170
- Shape the raw refusals on eval drafts, and raise the wall on swarm launch by @ignaciojimenezr in #5169
- ops: version the Convex surface-enumeration monitor by @chelojimenez in #5174
- Public model-lease route: let CI evals use MCPJam credits by @ignaciojimenezr in #5166
- Reset Describe chat on undo and simplify evaluator UI by @ignaciojimenezr in #5173
- [Judge 4/5] Align UI, SDK, CLI, and MCP grading contracts by @chelojimenez in #5102
- [Judge 5/5] Show effective automatic grading state by @chelojimenez in #5103
- Document cloud eval judge-backtest CLI command by @mintlify[bot] in #5179
- Point the inspector at the Convex custom domains (rt / rt-http) by @chelojimenez in #5181
- swarm: send the session index every persona turn is billed against by @nachocossio in #5184
- Fix credit purchase errors and guide Free users to Plans by @prathmeshpatel in #5185
- Docs: note free plan credit purchase restriction by @mintlify[bot] in #5191
- route error: show an access-denied screen instead of a stack trace (BB-250) by @SebasKoria in #5195
- feat: run SDK tests with latest saved clients and show version tooltips by @ignaciojimenezr in #5175
- Fix "Invalid or unsupported model" error for 25 BYOK models with bare model ids by @sophiemarinavo in #5148
- fix: match eval result filters to displayed verdicts by @ignaciojimenezr in #5199
- Let the evals action read a MCPJam behind Cloudflare Access by @ignaciojimenezr in #5150
- tokenizer: authenticate hosted backend requests server-side by @prathmeshpatel in #5203
- Remove the judge's paused state; grading is a suite setting by @chelojimenez in #5182
- fix: open SDK test cases in the current case workspace by @ignaciojimenezr in #5206
- Update CI-synced test case docs for read-only workspace by @mintlify[bot] in #5208
- Scorecard reason cleanup + Integrations tab behind a feature flag by @ignaciojimenezr in #5197
- Promote to test case lands in the Evaluate redesign by @ZeHuari in #5211
- Surface abuse containment and attest forwarded client IPs by @chelojimenez in #5178
- Clarify credit upsells and add frontier model sign-in modal by @prathmeshpatel in #5192
- Docs: update billing/BYOK dialogs and frontier sign-in by @mintlify[bot] in #5218
- fix: send all test-case links to Ding Dong by @ignaciojimenezr in #5210
- fix: show V2 organization usage as a remaining credit tank by @prathmeshpatel in #5187
- Disclose per-trial verification and consolidation sources on AI findings by @chelojimenez in #5221
- ops: move surface-enumeration monitor prose into rationale by @chelojimenez in #5224
- Fix swarm lifecycle, trace reporting, and widget browser races by @chelojimenez in #5223
- Let a model finish quietly after it has already used a tool by @ignaciojimenezr in #5217
- Apply pinned host styling to Swarm and Evaluate transcripts by @chelojimenez in #5226
- Preserve recorded swarm traces and report hosted turn failures by @chelojimenez in #5215
- Ground swarm sessions with target discovery and prerequisite setup by @chelojimenez in #5216
- Use host-styled inspector transcripts for User Testing and Sessions by @chelojimenez in #5229
- Reuse the shared host shell in Playground and live chat by @chelojimenez in #5232
- Unify transcript spacing and remove nested preview frames by @chelojimenez in #5233
- Unify transcript error and empty state presentation by @chelojimenez in #5234
- Run swarm grounding and setup in the background, with no UI by @chelojimenez in #5238
- Make the User Value Chain say what was expected and what happened by @chelojimenez in #5171
- Pass backend generation refusals through instead of answering 500 by @chelojimenez in #5180
- Say generation and insights are included, not billed to the customer by @chelojimenez in #5183
- Fix CI regressions from swarm findings and transcript updates by @chelojimenez in #5242
- Sync swarm grounding contract: drop people from facts, dedupe entities by @chelojimenez in #5239
- swarms: put the Run tab last on the run detail page by @nachocossio in #5196
- chore: release changeset for inspector, CLI, and SDK by @ignaciojimenezr in #5245
- Let people stop an eval run from every page that shows it running by @ignaciojimenezr in #5214
- Document cancel run button and comparison delta format by @mintlify[bot] in #5247
- Say why a model step came back empty, on hosted and BYOK chat by @chelojimenez in #5249
- ding dong by @ignaciojimenezr in #5240
- Persist per-step model requests across all Raw views by @chelojimenez in #5246
- Skip the evals action smoke while its repo vars are unset by @ignaciojimenezr in #5254
- Stop a swarm from the Run swarm step by @chelojimenez in #5256
- Report swarm execution, goal grading, and evidence independently by @chelojimenez in #5248
- Show verified finding recurrence and robust clustering controls by @chelojimenez in #5243
- Lead Findings with the suggested fix by @chelojimenez in #5260
- Release: wait out npm's 5-minute packument CDN cache before changesets publishes by @chelojimenez in #5263
Full Changelog: v3.6.0...v3.7.0