Changelog
New features, improvements, and fixes shipped to Aiqaramba.
0.43.0 September 9, 2026
new
Close and reopen findings from the API — Findings can now be resolved, dismissed, or reopened through the API, one at a time or up to a hundred in one atomic call. Every action needs a reason, and it is recorded in the finding's history together with who applied it and through which surface, so a scripted clean-up stays as auditable as a decision made in the app.
new
A status history on every finding — Each finding now keeps every change of state it has been through: who resolved, dismissed, reopened, or regressed it, from the app, the API, the chat, or the auditor after a retest, and why. The finding page shows the full trail, the API returns it with the finding, and the audit log lists each change as it happens.
new
A findings summary and a stale-finding filter — One call now returns a project's findings counted by status, severity, type, and source, so you can size a backlog before touching it. The findings list also gained the filters triage actually needs: not_seen_days for findings with no recorded observation during that period, a page URL fragment to sweep one screen, and the run that observed them. A missing recent observation does not prove that an issue was retested or fixed.
improved
Findings paginate properly — The findings list now reports the total across every page and hands back a cursor for the next one, so a large backlog can be walked in full instead of stopping at the first hundred.
new
Triage findings in chat — The chat assistant can now summarise a backlog, surface the findings worth closing with the evidence for each, and, once you agree, apply the closes with a stored reason. Propose, confirm, apply, without leaving the conversation.
0.42.0 August 16, 2026
improved
A sharper reviewer behind every run — Every finished run is reviewed by an auditor model before you see the verdict. We benchmarked that reviewer on a purpose-built suite of trap cases — runs that lie about succeeding, planted bugs, impossible instructions — and moved it to the configuration that judged them best. Verdicts now catch more agent mistakes and misreport less, at a lower cost per review.
new
Choose the auditor model per run — A run can now name the model that audits it, via auditor_model on the run endpoints — alongside the existing per-run auditor thinking control. Useful when a specific run deserves a deeper (or cheaper) review than the default.
new
Store sessions now, audit them later — Sync can now bank every session in a window without spending an audit — a "store only" mode alongside the usual friction-session review. Stored sessions show a "stored — not audited" state with an Audit button, and re-importing (or passing skip_audit on the API) controls whether a session is reviewed on the way in. It lets you keep a full baseline cheaply and audit selectively.
improved
Tighter authentication — Authentication now fails closed: a session or tenant that can't be verified because of a database error is rejected rather than let through, and the auth path was consolidated so every request resolves to a single verified identity. We also removed an experimental payment feature and its unauthenticated API surface.
0.41.0 August 5, 2026
new
Schedules that match how you actually release — A schedule can now fire on a monthly pattern — including offsets such as "the Monday before the last Thursday" — or just once, in the timezone you choose. The schedule page previews the coming fire times, so you can check a pattern before it runs.
new
Schedules can carry variables — A schedule can now hold variables that are filled in on every fire, alongside built-in values for the fire date and a unique fire id. Test plans that need a run tag or a date can finally go on a schedule.
new
Findings analytics on the dashboard — The dashboard now charts your active backlog by severity and by type, plus a 60-day histogram of findings found per day. Every chart segment links straight into the filtered findings list.
improved
A denser, more honest tests list — The tests list fits roughly twice as many tests per screen, and each test carries one health pill — healthy, flaky, failing, or not run yet — computed from its recent runs. Filters are now multi-select, and rows with an agent in flight pulse so you can see live work at a glance.
improved
A quicker start for new accounts — Signing up is now a short wizard — organisation name and timezone — that drops you straight onto the dashboard, where a needs-attention board and an activity feed guide your first steps.
improved
A simpler finding lifecycle — We removed the separate triage queue. A finding is now simply open, regressed, resolved, or dismissed — one list, fewer states, nothing to shepherd through a workbench.
0.40.0 July 29, 2026
improved
Your findings list shows only your product — Findings about our own test harness (agent behaviour, test-data artifacts) no longer appear in your findings list, the findings API, or run views. Your list is your product's defects, nothing else.
new
Discovery options in the CLI — The aiqa CLI now exposes discovery authentication, depth, and context flags, so scripted discoveries can use everything the API supports.
new
Agent egress auditing — We now maintain a fleet-wide ledger of every domain our agents contact, rolled up from run traces. Part of our ongoing safety and reliability work: it lets us verify agents stay where they belong and catch anything unexpected early.
improved
A rebuilt billing page — Billing and usage now tell you what your plan includes, what you have used this cycle, and exactly when and how much you will next be charged. The page matches the rest of the app and reads properly on a phone.
improved
Your agent allowance follows your billing date — The monthly agent allowance now renews on your billing date rather than on the first of the calendar month. If you pay yearly, your allowance still renews every month, on the same day of the month you subscribed.
improved
Storage breakdown opens straight away — The storage breakdown used to wait for every figure to be measured before it appeared. It now opens immediately and fills the numbers in as they arrive.
fixed
A cancelled subscription now says it is cancelled — Cancelling from the billing portal left the billing page describing an upcoming renewal. It now shows the date your subscription ends and confirms that nothing further will be charged.
fixed
Scheduled runs on the day the clocks go forward — A schedule set for a time that does not exist on the spring daylight-saving day fired an hour early. It now runs at the first moment that does exist, so a run is never brought forward and never skipped.
0.39.0 July 16, 2026
new
Test coverage overlay on the app map — The discovery app map can now overlay QA coverage, so you see at a glance which pages of your app are exercised by tests and which are blind spots.
new
Authenticator app 2FA support — Agents can now log in to apps protected by authenticator-based two-factor auth. A dedicated TOTP step generates the current code at the moment it is entered, and discoveries carry the 2FA setup through to every follow-up agent automatically.
new
Discovery drafts a test plan for you — Finishing a discovery on an authenticated app now produces a draft test plan with a login step up front, so the tests it proposes reuse one authenticated session instead of each logging in from scratch.
new
aiqa command-line tool — A new CLI for scripting the platform from your terminal, including a prose-based way to feed data into test-plan runs.
new
Waiting-agent notifications — You now get an event when an agent is waiting on something (for example an email that has not arrived), so stuck runs surface instead of silently idling. Agents can also opt in to an identifiable browser user agent so you can spot them in your own logs.
improved
Read-only JavaScript execution — The agent's in-page JavaScript tool is now strictly read-only: agents can inspect state but can no longer mutate your app through scripts, and they are held to interacting through the real interface with an honest wrap-up when they cannot.
improved
Automation run history — The automation detail page now lists recent runs, and manually triggered automations record their events correctly.
improved
Livelier plans and sharper traces — Active test plans now animate while running, and trace screenshots render more crisply.
fixed
Plan agents no longer stall on missing config — Run configuration is now carried into every step's browser lease, fixing test-plan agents that could stall waiting on settings that never arrived.
fixed
Triage on mobile — Unified the classification selectors in the triage modal and fixed its layout on small screens.
0.38.0 June 29, 2026
new
Advanced browser actions — A new low-level action tool gives agents precise control for interactions that plain clicks and typing cannot express: hovers, drag-and-drop, and key combinations.
new
Discovery agents stay logged in — Discovery agents now authenticate once, stay on your app instead of wandering off-site, and follow-up agents reuse the explorer's session rather than logging in again.
improved
Documentation revamp — The public docs got a full refresh: new look, rewritten content, built-in search, and clickable figures throughout the test-plan and writing-tests guides.
improved
Tests list usability — Entire rows on the tests list are now clickable, and the select-all checkbox works correctly.
improved
Experimental MCP server retired — The experimental MCP server has been removed. Chat, the web app, and the API remain the supported ways to drive the platform.
fixed
Clarification answers in the trace — When you answer an agent's clarification question, the answer now shows up in the trace, and duplicate answers are no longer recorded.
0.37.0 June 24, 2026
new
Findings triage workbench — A dedicated workbench for working through open findings one by one: a source banner for context, live stats that update as you go, and a split button to triage and immediately attempt a fix in one motion.
new
Per-test device selection — Choose the device a test runs on directly in the test form. Mobile testing is now clearly flagged as beta while it matures.
improved
Journeys are now Tests — "User journeys" are renamed to "Tests" across the whole product, with a restyled flat layout and general UI polish. Same concept, clearer name.
improved
Faster bulk finding actions — Bulk triage operations on findings are noticeably faster on projects with large finding counts.
improved
Self-hosted fonts and assets — Fonts and frontend assets are now served from the platform itself: faster loads and no third-party font requests.
fixed
Agents only see visible elements — The page snapshot agents work from is now filtered to genuinely visible elements, so they no longer try to interact with hidden or off-screen controls.
0.36.0 June 15, 2026
new
Attempt a fix — Launch a coding agent straight from a finding. Connect a GitHub repository to your project and the agent checks out the code, investigates the runs behind the finding, writes a fix, and opens a pull request that links back to the finding. You can add your own context to steer it, and re-attempt at any time.
new
Code review loop on fixes — Every proposed code fix is reviewed by a second, independent reviewer agent before the pull request is submitted. The reviewer's verdict and full trace are shown alongside the fix.
new
Code-change trace and diff viewer — Follow the coding agent step by step in the same trace viewer used for browser agents, with a diff browser showing exactly what changed: per-file expand and collapse, clean gutters, and context collapsed by default.
new
Test plan import and export — Export a test plan to a file and import it into another project, making plans portable across environments.
new
Automations — Scheduling has been revamped into Automations: pick any target, set a cron schedule, and manage everything from a reworked UI.
new
Per-run thinking control — Choose how much reasoning effort an agent spends on a specific run via the API, so quick smoke checks stay fast and hard flows get more deliberation.
improved
Independent auditor — The post-run auditor now runs on its own dedicated model, keeps its reasoning on record, no longer adopts the performing agent's identity, and reopens a resolved finding when an audit re-confirms the issue. The run phase is now called "auditing" to match.
fixed
Validation errors visible to agents — Form validation errors on registration pages were missing from what agents could see, causing them to miss real feedback from your app. They are now surfaced.
fixed
Trace page console errors — Cleaned up JavaScript errors that fired while the agent trace page loaded.
0.35.0 June 10, 2026
new
Shadow-DOM aware browsing — Agents can now see and interact with content inside shadow DOM, including closed shadow roots, and handle CAPTCHA challenges. A new find tool lets them locate elements that never appeared in the classic accessibility tree.
new
Agent and auditor reports side by side — The report modal now shows both the agent's own report and the independent auditor's report in tabs, so you can compare what the agent claimed with what the audit found.
new
Agent traffic tagged in your analytics — Agent browsers now mark themselves on every page they visit, so tools like PostHog can cleanly separate agent traffic from real users and your product metrics stay honest.
improved
Smarter discovery exploration — Discovery now explores by interactive affordance rather than by URL: it detects pointer-cursor elements as interactive, spawns follow-up agents for anything left unexplored, and spends its budget on the most product-relevant leads first.
improved
Chat briefing with release readiness — The chat briefing now opens with a release-readiness read on your project and a conversational summary instead of a wall of finding cards, and surfaces scheduling where relevant.
improved
Test-plan authoring guidance — The rules that make plans succeed (a login step up front, actor profiles, explicit verification) are now taught directly in chat, the docs, and the plan builder itself, based on an audit of real production runs.
improved
Runs only block on critical findings — Test-plan runs now only halt downstream steps on critical product findings; lower-severity observations are recorded without stopping the plan.
fixed
Clicks bound to the named element — Clicks are now bound to the element the agent actually named and fail closed when the page has changed underneath, instead of clicking whatever moved into that position.
fixed
Overlays and icon-only controls visible — Top-layer overlays such as modals and toasts now appear in what agents see, overlay text is surfaced, and icon-only buttons get labels so agents can identify them.
fixed
Bad URLs no longer restart the run — Navigating to an invalid URL is now reported to the agent as a normal error to recover from, instead of crashing and retrying the entire run.
0.34.0 May 29, 2026
new
Flag a finding as a design decision — When a reported issue is actually intentional behaviour, mark it as a design decision instead of a false positive. The finding closes and a design_decision product insight is recorded, so the synthesis and every future agent understand it — follow-up agents and the dedup/audit mechanism stop re-reporting it. Available as a single and bulk triage action, with its own status filter.
0.33.0 May 29, 2026
new
Reorderable plan variables — Plan variables can now be reordered with up/down controls in the settings drawer. The chosen order drives the variable pills, the Start-run form, and the drawer itself. Existing variables keep their current order.
new
Start-run autocomplete — Text fields in the Start-run dialog now suggest values you've used in previous runs of the same plan, newest first, as a dropdown.
new
Masked credential fields — Plan variables that look like secrets (password, token, secret, api_key, …) render as masked password inputs in the Start-run dialog and are excluded from autocomplete, so credentials never appear in a plaintext suggestion list.
new
Show/hide on secret fields — Masked Start-run fields now have an eye button to reveal or hide the value while typing.
new
Trace screenshot lightbox — Clicking a screenshot in the agent trace now opens it full-size in an overlay instead of a new browser tab.
new
Clipboard capture on click — Agents now record clipboard copies triggered by a click (e.g. a 'copy link' button), improving trace fidelity and downstream steps that reuse copied values.
improved
Lower agent token usage — Long agent runs cost less: browser_network_events is capped to the most recent entries, large tool outputs are masked once the turn that consumes them passes, and Gemini implicit prefix caching is enabled with a cached-token metric surfaced in telemetry.
improved
Agents pages redesign — Refreshed agents list and detail pages with a cleaner layout and status presentation.
improved
Live model settings — Thinking-level changes now take effect without a server restart. Gemini 3.5 Flash defaults to minimal thinking for faster, cheaper runs.
improved
Contextual back navigation — The '← Back' context (?from=) is now preserved when switching tabs on the agent page and when drilling into a trace step.
improved
Faster page loads — Static assets are served with immutable caching and scripts are deferred, reducing repeat-visit load times.
fixed
Stopped plan step no longer wedges the run — Manually stopping a test-plan step's agent now ends the whole run as stopped, instead of leaving it stuck in 'running' indefinitely.
fixed
Trace screenshot sizing — Stabilised screenshot dimensions and position in the trace view so frames no longer jump or stretch.
fixed
Click accuracy — Fixed click-rectangle capture that could cause occasional misclicks on re-rendering pages.
fixed
Drawer scrolling — Drawer bodies now scroll when their content overflows the viewport.
fixed
Project detail timestamp — Fixed raw '@Timestamp' text rendering on the project detail page.
fixed
Stale agent cleanup — A daily reaper now resolves agents stuck in waiting or summarizing states.
0.32.0 May 28, 2026
new
Finding discussions — Threaded discussions on individual findings, so you can comment, ask questions, and track resolution without leaving the finding.
new
Findings API — New read endpoints: GET a project's findings list and GET a single finding by ID.
new
Activity-pulse sparkline — Findings overview rows and the finding detail aside now show a recent-activity sparkline so you can spot which issues are actively moving.
new
Click overlay on trace screenshots — Trace screenshots now show a marker pinpointing exactly where the agent clicked, using sub-pixel coordinates.
new
Styled error pages — Web error pages (404, 500, …) now match the app's design instead of plain browser text.
new
Insight annotations — Annotate insights directly from chat, with a synthesis card that polls for generate/regenerate progress.
new
Android discoveries — Run discoveries against Android devices: a device toggle and app-artifact picker are available under Advanced settings on the discovery create form.
improved
Finding detail & overview revamp — Redesigned finding detail UI and polished the findings overview (checkpoint column dropped). A single 'False positive' action replaces the old 'Dismiss' and 'Not relevant' options.
improved
Integrations page — Reworked into a flat, full-width layout, with the GitHub integration slimmed down.
improved
Active project highlight — The current project is now highlighted in both the sidebar and the project selector.
improved
Findings filter — Added select-all / deselect-all, closes on click-outside, and defers submission until the panel is closed so you can set multiple filters at once.
improved
Test create & detail rebuild — Reworked the test creation and detail experience for a cleaner flow.
improved
Default model: Gemini 3.5 Flash — The default agent model is now Gemini 3.5 Flash (GA) — larger context and faster output.
fixed
Trace screenshot loading — The trace view now handles slow or brittle screenshot loads gracefully, with distinct UI states instead of broken images.
fixed
Self-eval severity gating — Agents only fail their self-evaluation on critical or high severity findings, not on low-severity observations.
0.29.0 May 22, 2026
new
Discovery v2: interactive app map — Redesigned discovery detail page with a two-panel layout. Left sidebar shows discovered pages grouped by type. Pannable/zoomable dagre canvas with clickable nodes. Selecting a page shows inline findings, test coverage with run history heatmaps, and page connections.
new
Discovery depth control — Choose quick (3 agents, ~5 min), standard (10 agents, ~15 min), or deep (25 agents, ~30 min) when starting a discovery. Available in the web form, API, MCP, and chat.
new
Discovery login credentials — Enter email and password when starting a discovery so agents can authenticate and explore pages behind login walls. Credentials propagate to all continuation agents automatically.
new
Three-state heatmap — Run result squares now show green (pass), amber (pass with non-blocking issues), or red (fail). The amber state surfaces medium-severity findings on runs that technically passed. Applies to dashboard, test detail, test plan runs, and discovery.
new
Collapsible main sidebar — Hamburger toggle in the app header collapses the navigation to an icon-only strip. State persists across page loads. Discovery pages auto-collapse the sidebar for more canvas space.
new
Contextual back navigation — Back links now adapt based on where you came from. Clicking an agent from a discovery shows '← Back to discovery' instead of '← Back to test runs'. Full page state (selected node, filters) is preserved in the URL.
new
Ref-based click verification — browser_click and browser_type now accept a ref parameter with the expected element text. If the page changed since the accessibility tree was read, the tool auto-corrects by searching nearby indices for the matching element.
new
Discovery agent summarization — Discovery agents now run the summarizer, producing per-agent findings that appear on the agent page and in the discovery findings section alongside the app-map-level findings.
improved
Evidence deep-linking — Finding evidence links now go directly to the trace step where the issue was observed (/agents/{id}/trace#step-N), not just the agent overview page.
improved
Finding cards on discovery page — Each finding shows type tag (bug, ux_issue, accessibility), source tag (environment, platform), relative timestamp, and differentiated links: 'View evidence' for trace-backed findings, 'View details' for findings without trace evidence.
improved
Discovery running state — Replaced the blank spinner with an informative scanning state showing the URL being explored, agent count, and a three-step progress indicator (Exploring → Building → Generating).
improved
Test naming — Removed tier prefix system (Access:, Functionality:, etc.) from all test naming across the product. Tests now use plain descriptive names.
improved
Discovery findings quality — buildFindingsPrompt now requests source, impact, and remediation per issue. BuildTestPromptTemplate includes steps. Findings are aligned with the summarizer schema.
fixed
Chat tool result display — Tool results with a message field now render as clean text instead of raw JSON in the chat interface.
fixed
Sidebar class collision — Discovery page sidebar CSS no longer overrides the main app sidebar header styling.
fixed
All lint issues resolved — Fixed gosec, errcheck, errorlint, nestif, and staticcheck issues across the entire codebase. Zero lint issues remaining.
0.28.0 May 20, 2026
improved
Personas renamed to Mailboxes — "Personas" are now called "Mailboxes" everywhere in the platform — navigation, chat help, docs, pre-flight checklists, and API references. The concept is the same (an email identity with credentials that agents use during tests), but the new name better reflects what they actually are.
0.27.0 May 14, 2026
new
Agentic summarizer — Replaced the single-shot 450KB conversation dump with a multi-turn audit loop. The summarizer uses trace step tools to investigate specific failures on demand, eliminating Gemini 504 timeouts on long-running agents.
new
Recording to Test Plan workflow — Upload a demo video or record with the browser extension. AI extracts test flows, credentials, required files, and a transcript. Review flows, accept to create tests, then assemble into a test plan.
new
Docs search — Fuzzy search across all API endpoints, guides, and use cases. Powered by Fuse.js with keyboard navigation and '/' shortcut.
new
Recording to Test Plan guide — New use case guide covering both paths (extension + video upload), flow review, and test plan assembly.
new
Mic narration in extension — Record microphone audio alongside DOM events using standard getUserMedia. Audio is stored to GCS for AI analysis.
new
Test plan status legend — Info button on the test plan run page explains all 8 node states and how they affect downstream steps.
improved
Video upload page — Moved into the platform UI with XHR progress tracking, clear file-attached visual state, and navigate-away warning.
improved
Extension download page — Integrated into the platform with download cards, installation steps, and recording guide.
improved
Recording detail page — Shows all linked tests (not just one), analysis polling with spinner, flow cards with accept/dismiss, setup requirements as collapsible sections, and transcript panel.
improved
Test plan scroll preservation — Switched polling from outerHTML to morph swap so horizontal scroll position is preserved during status updates.
fixed
Skip cascade propagation — Test plan coordinator now propagates skips through the entire dependency chain in a single reconcile call, not one level at a time.
fixed
Browser URL context for shared sessions — Agents reattaching to a shared browser session now receive the current URL, preventing navigation to hallucinated URLs.
fixed
Extension re-recording — Fixed 0-event bug on second recording without page reload. Content script teardown guard prevents duplicate message listeners.
new
Mobile agent support — emulatord service enables agents to test Android apps via emulated devices with self-signed APK URLs.
new
Structured LLM errors — Replaced error string matching with a typed ProviderError, making retry logic and error reporting more reliable across LLM providers.
new
Tenant-tagged logging — Every log line now includes tenant_id and tenant_name for easier debugging in shared infrastructure.
new
aqlog search utility — Quick log search tool deployed to VMs for faster production debugging.
improved
Test plan node-run status — Node-run status now mirrors the agent's QA verdict (not just completion status), so failed QA shows as failed in the plan view.
improved
Discovery insight cap — Raised the per-discovery insight limit from 5 to 25.
0.26.0 May 13, 2026
fixed
Strict JSON schemas for LLM responses — Constrained one-shot LLM calls with strict response schemas, fixing serialization failures in summarization.
fixed
Summarize worker bugs — Fixed three bugs that blocked test plan completion: issues-found mismatch, missing completion report extraction, and stale message references.
fixed
Silent fallback removed — Removed the degraded summary fallback that silently hid errors, restoring visibility into stuck summarization states.
0.25.0 May 12, 2026
new
Test plan variables and UI improvements — Plan-level variables, orchestrator concurrency fixes, and step editor improvements for browser sharing and persona overrides.
improved
Agent text extraction — Correctly extract text from anchor tags with nested elements and labels, improving accessibility tree accuracy.
improved
Severity-aware success override — Don't override agent success for low-severity observations — only genuine issues flip the outcome to failure.
improved
Dependency updates — Updated River, Anthropic SDK, PostHog, templ, html-to-markdown, Google API, and Gemini SDK.
fixed
Video upload 403 — Fixed GCS signed URL permissions and recordings stuck in uploading state.
0.24.0 May 8, 2026
new
Test plans — DAG-based test plans with shared browser profiles, typed inputs/outputs, dependency edges, and a visual run viewer.
new
Browser recorder extension — Chrome and Firefox extension that captures rrweb DOM events and optional video/audio for automated test test generation.
new
Toast notifications for billing — Replaced blank 402 pages with non-intrusive top-right toast notifications for quota and billing limits.
improved
Recordings UI — Redesigned recordings list to match app table style, added content script on-demand injection and SPA navigation re-injection.
fixed
Usage calculations — Exclude trace data from usage counts and anchor billing to subscription period instead of calendar month.
fixed
Replay player auth — Events endpoint now served via session-authed web route so the replay player works without an API key.
0.23.0 May 5, 2026
new
Sub-agents — Agents can spawn sub-agents for multi-agent coordination — e.g. user A invites user B, then user B accepts. Each sub-agent gets its own persona, credentials, and browser session. The parent agent orchestrates and waits for results.
new
Bulk run tests — Run multiple user tests at once from the tests list page. Select the ones you want and launch them in a single action.
new
Chat file attachments — Share files (PDFs, images, text documents) directly in the chat. Uploaded files are injected into the conversation so the assistant can reference them when creating tests or analyzing your app.
new
Chat URL fetching — Paste a URL in the chat and the assistant fetches the page content, converts it to markdown, and uses it as context. Great for sharing documentation or landing pages.
new
Chat file browsing — The assistant can browse and read files from your tenant file storage.
new
Update project from chat — Set entry URL, context notes, and other project settings directly through the chat assistant.
new
Benchmark flag — Support users can flag difficult test runs as benchmarks for model comparison analysis.
new
Live welcome panel — After onboarding, the welcome page shows a live panel with your first test run's progress and results.
improved
Expandable tool calls in chat — Completed tool calls now show a clickable chevron that reveals the raw JSON result — see exactly what happened behind the scenes.
improved
Chat capabilities awareness — The system prompt now describes platform capabilities at the product level. The assistant discovers scheduling, regression, multi-agent, and other features automatically.
improved
Browser session persistence — Browser sessions survive agent restarts via a new Lease/Manager system. Degraded reattach injects a warning so agents don't report platform events as findings.
improved
PDF and document support — Both Claude and Gemini providers now support native document blocks (PDFs), not just images.
fixed
Trace screenshot cropping — Fixed screenshot cropping on the trace view by replacing flex centering with margin auto.
0.22.0 April 30, 2026
new
Entry URL in project settings — Set the starting page for test runs directly from the project settings form.
improved
Regression auto-test relaxed — Relaxed the auto-test gate, surfaced regression status, and always include run ID.
fixed
Trace fallback removed — Removed mid-flight trace fallback heuristic that could show stale data.
fixed
Trace header jitter — Constrained trace detail header height to prevent layout shift from large screenshots.
0.21.0 April 28, 2026
new
Billing & Usage page — Dedicated page with tier-coded card, plan limits, progress bars, and support-override pill.
new
Differential analysis — Agent summarization pipeline now includes differential analysis comparing runs against prior results.
new
Trace findings UI — Distinguish errors from warnings, collapse long finding lists, and severity-coloured tab dot.
new
Screenshot carry-forward — Trace viewer carries screenshots forward across steps so context is never lost.
new
Findings linked to trace steps — Findings now reference the specific trace step where they were observed.
fixed
Raw tab access control — Agent raw tab restricted to support users only.
fixed
Findings dedup & auto-resolution — Improved dedup accuracy, scoped auto-resolution to correct project, cleaned up evidence display.
0.20.0 April 27, 2026
new
Custom browser headers — Project-level custom headers injected into all browser requests, enabling auth-gated or token-protected test targets.
improved
Stronger QA incentives — Todo-list incentives reinforced so agents complete all checklist items before finishing.
fixed
Digest findings link — Fixed 404 on findings link in digest emails.
0.19.0 April 21, 2026
improved
Discovery boolean — Replaced agent_type enum with is_discovery boolean for simpler agent classification.
new
Project favorites — Per-user project favorites for quick access from the sidebar.
new
Welcome page — First-login users see a welcome page with setup and non-setup modes.
improved
Inline rename validation — Rename inputs validate and surface restrictions inline before submit.
improved
Tenant name in digests — Digest emails now include the tenant name for sender clarity.
0.18.0 April 15, 2026
new
Test creation revamp — Guided flow with manual form and AI generation prompt for creating tests.
new
Auth page redesign — Minimal centered layout for login and registration screens.
new
Quick impersonate — Support tenant detail page now has a one-click impersonate button.
fixed
Findings dropdown clipping — Actions dropdown no longer clips on findings page.
fixed
Real digest preview data — Digest preview uses real schedule data instead of hardcoded samples.
0.17.0 April 14, 2026
new
Global project selector — Sidebar command palette for switching between projects.
new
Documentation pages rework — Reworked in-app docs with drawer navigation.
new
Onboarding flow — Post-auth onboarding page with auto-provisioning, timezone detection, schedule seeding, and first-run trigger.
new
Regression auto-test — Automatic test creation when no existing test matches a regression trigger.
new
Chat overlay drawer — Replaced chat sidebar with overlay drawer.
improved
Users management redesign — Modern UI for the users management page.
improved
Sidebar restructure — Collapsible sections, Settings renamed to Setup.
fixed
Schedule bugs — Fixed 5 scheduling bugs: edit no longer wipes run history, Slack suppression for schedule-sourced agents.
fixed
Onboarding fixes — Credentials use external type, compact layout for small screens, project cookie handling.
0.16.0 April 9, 2026
new
Vertex AI Priority PayGo — Per-tenant toggle routes Gemini requests through Google's priority compute queue, reducing 429 rate-limit failures on preview models.
improved
Findings-first digest — Schedule digest emails now lead with findings and pull CTA buttons forward for faster triage.
fixed
Orphaned tenant prevention — Deleting the last user on a tenant is now blocked to prevent unrecoverable orphaned tenants.
0.15.0 April 9, 2026
new
Discovery findings ingestion — Site discoveries now produce structured findings with deduplication and evidence linking.
new
Manual Send Digest button — Re-queue a digest from the schedule detail page when the automatic one fails or times out.
new
VAT/tax ID collection at checkout — Stripe checkout now collects VAT or tax ID during subscription signup.
improved
Findings-led chatbot — The chatbot now leads with findings instead of test run pass/fail stats.
improved
Findings dashboard source visibility — New source column and actionability guidance banners help distinguish customer-actionable findings from platform issues.
fixed
Schedule digest not firing — Both the API and MCP trigger handlers were missing the RiverClient reference.
fixed
Schedule edit form losing time value — Postgres stores run_at as HH:MM:SS but the time input expects HH:MM — values are now normalized.
0.14.0 April 9, 2026
new
Discovery findings ingestion — Site discoveries now produce structured findings. The LLM health assessment is parsed and each issue is upserted into the findings table with deduplication and evidence linking.
new
Manual Send Digest button — Re-queue a digest from the schedule detail page when the automatic one fails or times out.
new
VAT/tax ID collection at checkout — Stripe checkout now collects VAT or tax ID during subscription signup.
improved
Findings-led chatbot — The chatbot now leads with findings — bugs, UX issues, and severity breakdowns — instead of test run pass/fail stats. Briefing cards show individual findings and the system prompt is simplified.
improved
Findings dashboard source visibility — Default severity filter changed to all severities. New source column and actionability guidance banners help distinguish customer-actionable findings from internal platform issues.
improved
Clickable findings on agent detail — Findings listed on the agent detail page now link directly to the finding detail view.
improved
Pricing page refresh — Updated terminology to use test runs and user tests, added VAT exclusion note, and pointed trial CTAs to the in-app pricing page.
fixed
Schedule digest not firing on manual triggers — Both the API and MCP trigger handlers were missing the RiverClient reference, causing digest jobs to silently skip.
fixed
Schedule edit form losing time value — Postgres stores run_at as HH:MM:SS but the time input expects HH:MM — values are now normalized before rendering.
fixed
Finding evidence badge label — The evidence section on finding detail now shows "issues found" instead of the raw "completed" status for agents that finished with failures.
fixed
Severity pill color mismatch — Medium and low severity colors on the home screen now match the agent detail page.
fixed
Landing page nav links — All sub-pages now show the full navigation with consistent spacing and active-state highlighting.
fixed
Digest worker timeout — Increased the digest wait window from 30 minutes to 2 hours so long-running agent batches finish before the digest is generated.
0.13.0 April 7, 2026
new
Findings overview — The home page is now a dedicated findings view — see open issues, severity breakdown, and resolution status at a glance. Supports table and card layouts, multi-severity filtering, and project scoping.
new
Finding detail page — Click any finding to see its full context: what happened, impact assessment, remediation steps, the agent that discovered it, and a link to the recording.
new
Resolve, dismiss, and reopen — Triage findings directly from the detail page or in bulk. Mark issues as resolved, dismiss false positives, or reopen if they regress.
new
Bundled retest — Select multiple findings and launch a single verification agent that re-checks all of them, updating their status based on results.
new
Digest severity filter — Configure a minimum severity on schedule digests — receive only critical, high+critical, or medium+ findings in your inbox. Set it on the schedule create/edit form.
new
Handle native dialogs — New browser_handle_dialog tool lets agents interact with native JavaScript alert(), confirm(), and prompt() dialogs.
new
Support favorites — Support users can favorite tenants for quick access from the support dashboard.
improved
Findings in digest notifications — Schedule digest emails and Slack messages now include structured findings with severity badges, not just pass/fail summaries.
improved
Structured findings in agent summaries — Agents now produce machine-readable findings with severity, type, impact, and remediation — powering the entire findings workflow.
improved
Findings banner — The summary banner highlights what's going well alongside what needs attention, with clickable severity cards that filter the list.
improved
Empty states — Polished empty states across the app using consistent icon + title + action pattern.
fixed
Video and activity log sync — Fixed a mismatch where activity log entries didn't align with video timestamps after session reattachment.
fixed
iframe element resolution — Agents can now correctly identify and interact with elements inside iframes, with proper cross-origin annotations.
fixed
Element cache between iterations — Fixed a bug where the element cache was lost between agent loop iterations, causing stale element references.
fixed
Duplicate Slack notifications — Deduplication by channel target prevents the same notification from being sent multiple times.
0.12.0 April 5, 2026
new
Automatic database migrations — Migrations now run automatically on server startup using golang-migrate. No more manual migration steps during deploys.
new
Support user impersonation — Support users can view any tenant's data from a dedicated dashboard with search and favorites.
new
Interactive briefing cards — Chat briefings now display as grouped cards by test, making it easier to scan results.
new
Failure source classification — Agents classify failure root causes (app bug, auth issue, agent error) for better triage.
improved
Terminology rename — User-facing labels updated: agents → test runs, tests → user tests, discoveries → site discovery.
improved
Default max iterations raised to 200 — Agents now have more room to complete complex flows before hitting the iteration limit.
fixed
External credential form — Fixed duplicate input names causing wrong fields to submit for external credentials.
fixed
File copy on rerun and schedule trigger — Test files are now correctly copied to the agent workspace across all run paths.
0.11.0 April 2, 2026
new
Live progress checklist — Agents display a real-time progress checklist during execution, showing which steps are done, in progress, or pending.
new
Auth guidance in agent prompts — Agents receive authentication and registration guidance in their system prompt, improving login success rates.
new
Schedule edit page — Edit existing schedules with a pre-populated form — change tests, run time, days, and fallback persona/role.
new
Report bug from chat — New report_bug chat tool lets users file issues directly from the conversation.
new
Magic link and external credential types — New credential types for magic-link auth flows and external service accounts.
new
Agent source display — See whether a test run was triggered manually, by a schedule, or by a monitor.
improved
Retry as clone-based rerun — Re-running an agent now creates a clean clone preserving history, instead of mutating the original.
improved
Schedule test search — Search filter added to test selectors on schedule create and edit forms.
improved
PostHog SDK — Replaced custom PostHog wrapper with the official Go SDK for better reliability.
fixed
Agent stuck in pending after rerun — Fixed River job deduplication causing reruns to silently fail.
fixed
Stopped agents reverting status — Prevented stopped agents from being overwritten back to running or completed.
fixed
Chat persistence on disconnect — Chat conversation is now preserved when the client disconnects mid-stream.
0.10.1 April 1, 2026
new
Recurring schedules — Define a pool of tests and have the system automatically rotate through them on a configurable interval. Set a coverage window (e.g. 5 days) and daily budget — the scheduler picks the most overdue tests each tick, respecting your agent quota.
new
Schedule management UI — New Schedules page in the sidebar with list, detail, and create views. Add/remove tests, pause/resume, trigger manually, and view run history.
new
Schedule API — Full REST API for schedules: create, list, get, update, delete, manage items, view runs, and manual trigger. Plus 5 MCP tools for chat-based schedule management.
0.10.0 March 30, 2026
new
PM-friendly chat experience — The chat agent now acts as a quality partner for product managers. Describe flows in plain language and the system auto-generates test tests — no technical prompt syntax required.
new
Generate test from description — New generate_test tool translates plain-language flow descriptions into structured test tests with steps, importance, and agent prompts.
new
Coverage map — Ask "Are we ready to release?" and get a tier-grouped coverage breakdown with pass/fail/untested counts and a clear release recommendation.
new
Regression detection in briefings — The chat proactively flags flows that were passing but are now failing, with actionable recommendations.
new
Update test via chat — Change a flow description in chat and the underlying test prompt auto-regenerates to match.
improved
Proactive briefings — Chat now surfaces coverage gaps, stale tests, and untested flows — not just failures.
improved
Suggestion chips — New PM-relevant quick actions: "Are we ready to release?", "Describe a new flow to test".
new
Agent error classification — Agents now classify errors with structured error_classification and error_explanation fields for better failure analysis.
0.9.0 March 27, 2026
new
Agent memory — Project-scoped operational knowledge that agents save and recall across runs. Workarounds, efficient paths, and timing tips persist between sessions.
new
Background monitoring — The chat agent can launch background monitoring sessions — run all tests, poll for completion, analyze failures, retry with prompt fixes, and save learnings automatically.
new
Guided test creation — Describe a flow in plain language and the system generates a structured test test with steps and agent prompt. Available via /tests/guided.
improved
Chat overhaul — JIT briefings, smart suggestion chips, and copy button on assistant messages.
improved
Onboarding flow — Value-first onboarding: app URL → context notes → first test → aha moment, before asking about integrations.
improved
Delayed notifications — Slack and email notifications now fire after summarization so they include the evaluation outcome.
improved
tom-select / selectize support — Agents can now interact with custom dropdown libraries (tom-select, selectize) commonly used in web apps.
improved
Video skip-idle — Configurable idle threshold in video player — skip long waits in agent recordings.
fixed
Entry URL routing — Test entry_url now correctly sets BrowserURL so agents start at the right page instead of about:blank.
fixed
Context notes clear button — Removed native confirm dialog that blocked the HTMX clear request.
fixed
Agent success reconciliation — Agent pass/fail now reconciles with server-side summary evaluation for consistent results.
0.8.0 March 26, 2026
new
AI chat agent — A conversational AI assistant on every page. Ask questions, run tests, trigger discoveries, check health, and manage projects — all through natural language. Includes streaming responses, conversation history, and rich response cards.
new
Proactive status briefings — The chat agent greets you with a summary of what needs attention — failing tests, stale tests, and overall health.
new
Clickable suggestion chips — Context-aware action buttons appear based on the page you're on. One click to summarize an agent, run all tests, or check integration status.
new
Project context notes — Add free-text notes to any project (app URL, credentials, key flows). The chat agent uses these to give better answers.
new
Persona and role management — Personas are now separate from roles. Manage credentials per persona, assign personas to tests, and configure role-based access.
new
Announcements — Support team can publish banners to customer tenants for maintenance windows, feature announcements, or status updates.
improved
Chat-first dashboard — The home page is now centered around the chat agent with a project selector and quick link to the health board.
improved
Backward pagination — All list pages now support Previous and Next navigation.
improved
Default project on sign-up — New tenants start with a project and persona already created, so you can get started immediately.
fixed
Agent video playback — Fixed 404 errors for older agent recordings and improved video processing reliability.
fixed
Discovery credential propagation — Discovery credentials now correctly flow through to all continuation agents.
0.7.0 March 23, 2026
new
GitHub issue sync — Import GitHub issues as test tests filtered by label and state. Mirrors the full Linear sync workflow: automatic feedback comments, close-on-pass, reopen-on-fail.
new
GitHub webhook support — GitHub App webhook handler triggers automatic re-sync when issues are created, edited, or re-labeled.
new
GitHub auto-issue from integrations page — Configure per-project auto-issue creation targeting a GitHub repository, directly from the integrations page.
new
Agent video player — Watch recorded agent sessions with chapter markers and thumbnail previews powered by ffmpeg scene detection.
improved
Generic issue state columns — Renamed linear_state to issue_state so both GitHub and Linear synced tests display provider state in the tests list.
0.6.0 March 22, 2026
new
Slack integration — Connect Slack to receive notifications when agents start, complete, or fail. Configure per-channel rules with granular event filters.
new
Configurable issue destination — Choose whether failed agents create issues on GitHub or Linear, configured per project.
new
Multiple sync configs per Linear integration — Set up multiple label-to-project sync rules so different Linear labels route to different Aiqaramba projects.
improved
Support view shows tenant name — The support banner on agent detail pages now displays which tenant the agent belongs to.
improved
LLM-Ready Reference moved above the fold — The copy-for-LLM button on the docs page is now visible without scrolling.
0.5.0 March 20, 2026
new
Automatic Linear sync via webhooks — Linear issues with a configured label are automatically synced as test tests in real time. No manual sync needed.
new
CI/CD integration guide — New docs page walking through the full Linear + GitHub + regression testing setup.
new
Source filter on tests list — Filter tests by source (manual, discovery, linear, guided) to find what you need faster.
improved
Deploy health check in CI — Staging deploys now wait for the server to be healthy before running regression tests, preventing false positives.
fixed
API query param consistency — List endpoints now accept both project_id and project as query parameters.
0.4.0 March 17, 2026
new
MCP server built into the main binary — The MCP server is now part of the main Aiqaramba server. Connect your coding agent with a single command.
new
Linear OAuth integration — Connect your Linear workspace from the integrations page. Sync issues as test tests and post agent results back as comments.
improved
Discovery completion reliability — Fixed issues where discoveries could get stuck in a running state.
fixed
Email inbox tool link extraction — The email_read_inbox tool now correctly extracts links from HTML email bodies.
0.3.0 March 14, 2026
new
Guided test creation — A step-by-step wizard for creating test tests without writing prompts manually.
new
How It Works docs page — New documentation explaining Aiqaramba's feedback loops and testing philosophy.
improved
Project health dashboard performance — Reduced dashboard load time from 4 seconds to under 500ms by fixing N+1 queries.
improved
Onboarding flow — Updated onboarding to use guided test creation for a smoother first experience.
fixed
UI regressions — Fixed modal close behavior, docs redirect, discovery cards, and credentials navigation.
0.2.0 March 14, 2026
new
File management — Upload, browse, and attach files to agents. Files are stored per-tenant with storage quotas and progress tracking.
new
Test health board — Visual dashboard showing test pass/fail trends, heatmaps, and quick-run buttons.
new
Getting started widget — In-app onboarding guide for new users with step-by-step setup instructions.
improved
OpenTelemetry tracing — Replaced long-running parent spans with short-lived lifecycle spans for cleaner traces.
0.1.0 March 12, 2026
new
Initial release — Core platform with autonomous browser agents, project management, persona system, test templates, discovery engine, and GitHub integration.