CHANGELOG · SHIPPED WEEKLY

What we shipped.

No batch holds. No marketing windows. Every user-visible change lands the week it's written.

releases
15
user-visible changes
162
last shipped
Jun 17
JUNE 2026
  1. v0.11.0
    latest

    Evaluations and Agent Judge — score every agent run

    • NEWEvaluations scores agent runs after they finish — point a rule at any trace, session, agent, dataset, or time range
    • NEWNine zero-config rules grade task completion, error rate, latency, cost, hallucination, and more
    • NEWAgent Judge grades multi-file agent outputs — a fresh agent reads the files against your rubric and shows its work
    • NEWDeterministic prechecks gate every Agent Judge run, so structural failures fail fast with no LLM spent
    • NEWWrite custom eval rules as an LLM prompt, your own Python, or an agentic judge
    • NEWCompare two eval runs side by side and see exactly which verdicts flipped
    • NEWRun evals straight from the SDK — kick off a run, poll, and pull verdicts in a few lines
    • IMPGuard and Eval rules now share one Rules page — promote an eval rule to a real-time block in a click
    • IMPNew agent pages surface recent verdicts, diagnoses, traces, and drift in one place
    • IMPUpload agent-run file bundles to a dataset and browse them in the dashboard
    • IMPAdded a team tier to pricing
    • PRFEval LLM judges batch per trace, cutting judge calls by an order of magnitude
    • SECEval routes are gated by per-action permissions and plan limits
    • DOCNew eval docs cover runs, verdicts, compare, and Agent Judge

    +55 under-the-hood changes shipped

APRIL 2026
  1. v0.10.0

    Monochrome redesign and interactive product demo land

    • NEWInteractive four-tab product demo on the landing page — trace, conversation, guard, dataset
    • IMPMonochrome redesign across the marketing site and the dashboard
    • IMPDashboard chrome rebuilt — sidebar, topbar, home, onboarding, and settings rail
    • IMPConversations reimagined as a terminal-replay transcript with j/k navigation and / find
    • IMPObservability analytics laid out as a dense bento of stats and charts
    • IMPProduct, about, changelog, legal, and investor pages rebuilt on the new system
    • FIXPath traversal Guard rule tuned to drop false positives on normal project paths
    • FIXLLM judges only see prior violations when a rule opts in — less cross-rule noise

    +2 under-the-hood changes shipped

  2. v0.9.1

    SDK auto-grouping; sixteen new Guard rules

    • NEWTool-loop state machine stitches multi-turn tool calls into a single trace for Anthropic and OpenAI
    • NEWZero-decorator auto-grouping — stack-frame anchoring fuses sibling and nested LLM calls under one trace
    • NEW`client.chat.completions.parse()` structured-output calls now traced
    • IMPSelect all rules in Guard's attach-rules dialog in one click — no more one-by-one
    • IMPCmd-click now opens table rows, detail panels, and CTAs in a new tab
    • IMPAgent names show on the monitoring and traces tables instead of raw IDs
    • IMPInteractive CLI wizard adds default hints, TTY detection, and env-var pre-fill
    • PRFConversation detail loads all spans in a single batched call — no more N+1 fetches
    • SECTwelve new static Guard rules cover SQL injection, SSRF, path traversal, dangerous tools, and cost caps
    • SECFour new LLM judges catch toxicity, goal deviation, false completions, and RAG factual drift

    +7 under-the-hood changes shipped

  3. v0.9.0

    Guard now blocks LLM tool calls, full docs rewrite ships

    • NEWGuard blocks tool calls inside OpenAI and Anthropic responses — not just observes them
    • NEWPick fail-open or fail-closed per Guard call, or default it process-wide
    • NEWCodex agents now run through Guard with full enforce and audit modes
    • NEWTrace detail renders Guard violations as a rich panel — rules, action, modified input
    • IMPSelf-Heal usage broken out into its own tab on the usage page
    • IMPUsage page rebuilt with a monthly traces hero and cleaner per-resource rows
    • IMPSDK transport adds retries and backoff so traces survive flaky networks
    • DOCFull SDK and platform docs rewrite — every signature grounded in source code
    • DOCRenamed 'session' to 'conversation' across docs to match the SDK and dashboard

    +21 under-the-hood changes shipped

  4. v0.8.1

    Self-Heal gets depth control, live streaming, and Guard coverage

    • NEWSelf-Heal runs stream live progress — watch diagnoses unfold in real time
    • NEWPick diagnosis depth before kicking off a heal run — quick or deep
    • NEWCancel a running Self-Heal session mid-flight without losing the rest
    • IMPSelf-Heal diagnoses now execute in isolated sandboxes by default
    • SECGuard now watches Self-Heal's internal agents for cost, context, and dangerous ops
    • DOCNew Self-Heal docs — overview, deep runs, GitHub setup, VCS tracking

    +14 under-the-hood changes shipped

  5. v0.8.0

    The loop closes: Self-Heal and Datasets land

    • NEWNew product: Self-Heal — diagnoses failed agent runs and opens fix PRs in your repo
    • NEWSandboxed reruns replay each failure in isolation to nail the exact root cause
    • NEWConnect GitHub once — Self-Heal targets fixes at the agent code that broke
    • NEWNew product: Datasets — turn any trace or conversation into a labeled test set
    • NEWSpreadsheet view to edit dataset entries with folders, splits, and CSV import
    • NEWSDK ships `staso.dataset` — create, version, and evaluate datasets in code
    • NEWGenerate synthetic dataset entries to expand coverage from a handful of examples
    • NEWEvery trace records the exact git commit that produced it
    • NEWGitHub integration page in settings — one-click install per workspace
    • IMPSpan detail view shows source location and code-level context inline

    +38 under-the-hood changes shipped

  6. v0.7.0

    Guard: stop rogue agents before they act

    • BRKBreaking: `agent_id` renamed to `agent_name` across SDK and docs
    • NEWNew product: Guard — every tool call checked against safety rules in real-time
    • NEWAudit mode watches silently; enforce mode blocks dangerous calls before they execute
    • SECEight built-in rules ship day one — prompt injection, data exfiltration, jailbreaks, and more
    • NEWPolicies scope rules to specific agents, tools, or environments
    • NEWGuard monitoring: violation timelines, per-rule latency, agent risk breakdowns
    • NEWGuard works via SDK — one function call to protect any agent
    • NEWNew `/product/guard` marketing page with gallery and scene walkthroughs
    • NEWUsage-limit banner warns before you hit plan caps
    • NEWConversation-level analytics with per-session cost and token breakdowns
    • IMPSetup guide walks new users from zero to first trace in minutes
    • IMPTraces display agent names instead of opaque IDs
    • IMPOpenAI traces render with provider-aware pretty cards
    • IMPResizable trace panels remember your preferred layout
    • IMPInline copy-id fields on organization and workspace pages
    • FIXPII redaction covers additional edge cases
    • FIXCodex hook tracing stabilized across session types

    +35 under-the-hood changes shipped

  7. v0.6.0

    Codex tracing, pretty engine, and usage dashboard

    • NEWTrace OpenAI Codex agents with drop-in SDK hooks
    • NEWTraces render as rich provider-aware cards — Claude Code and Codex each get tailored views
    • NEWUsage dashboard rebuilt with per-org breakdowns and progress bars
    • NEWNew `staso sync` CLI command auto-configures hooks in one step
    • NEWNew interactive setup wizard — arrow-key driven SDK onboarding
    • NEW`staso uninstall`, `update`, `status`, and `version` CLI commands
    • NEW`STASO_AUTO_SYNC` env var for automatic config sync on every hook run
    • IMPChangelog page redesigned with a visual timeline
    • IMPTrace charts and stats bar overhauled with cleaner layout
    • IMPSpan tree gets color palettes and improved node rendering
    • IMPTrace search bar refined for faster, more intuitive filtering
    • IMPLanding page integration section updated with Codex
    • IMPCost breakdown simplified across trace and conversation views
    • SECPII automatically redacted from trace data on ingestion
    • DOCPrivacy policy and terms of service expanded

    +47 under-the-hood changes shipped

  8. v0.5.0

    Pretty trace cards and scoped search

    • NEWTrace spans render as rich visual cards — diffs, code, and tools at a glance
    • NEWSearch traces by specific field: name, input, output, or errors
    • IMPRepeated tool calls auto-collapse into tidy grouped rows
    • IMPLanding page adds problem, testimonials, and why-Staso sections

    +15 under-the-hood changes shipped

MARCH 2026
  1. v0.4.0

    Live streaming traces and shareable links

    • NEWTraces stream in real-time — see new data as it arrives
    • NEWShare any trace with your team via direct link
    • NEWNew timeline view shows how trace spans connect
    • NEWSystem-health indicator in the topbar with a link to the status page
    • IMPTrace viewer auto-scrolls to keep you at the latest content
    • IMPEnhanced date-time picker with more precise range controls
    • FIXWebSocket accept-then-authenticate fix prevents stream drops
    • FIXFilters correctly combine when using multiple criteria
    • FIXChart timestamps now display in your local timezone
    • FIXFilter panel scrolls properly on smaller viewports

    +18 under-the-hood changes shipped

  2. v0.3.1

    Smoother tables and sharper landing page

    • NEWInvestor-preview page with time-locked demo access
    • IMPTraces and conversations load continuously with infinite scroll
    • IMPCopy any trace or conversation ID in one click
    • IMPRefresh-token rotation keeps long dashboard sessions alive
    • IMPProduct screenshots expand to fullscreen on click
    • IMPLanding page hero refreshed with clearer product positioning

    +10 under-the-hood changes shipped

  3. v0.3.0

    Smarter monitoring and full workspace isolation

    • NEWNew monitoring dashboard with health status, sparklines, and unified controls
    • NEWCustom alert thresholds for error rate, latency, tokens, and cost
    • NEWAdjust chart granularity (5m to 1d) from the control bar
    • NEWKeyboard shortcuts: arrow keys navigate spans and detail tabs
    • NEWTraces open with first span selected, output tab shown by default
    • IMPSofter dark mode contrast for comfortable long sessions
    • IMPCode examples on the landing page adapt to your theme
    • IMPTop Spans table is expanded by default with sortable columns
    • FIXWorkspaces are now fully isolated across all metrics and charts
    • FIXAuto-refresh updates data without resetting your scroll or selection
    • FIXWorkspace switching correctly resets filters and reloads data
    • FIXRedis connection pooling fix eliminates prod stalls under load

    +16 under-the-hood changes shipped

  4. v0.2.0

    Claude Code tracing, dashboard overhaul, and SDK docs

    • NEWClaude Code integration auto-traces sessions via SDK hooks
    • NEWAuto-refresh control for live trace streaming
    • NEWCustom date range picker across all dashboards
    • NEWMobile-responsive sidebar for the dashboard
    • NEWOnboarding flow now tracks completion state per user
    • NEWAutomated email notifications via new cron service
    • NEWRun SDK directly with `python -m staso` in any venv
    • NEWSubscriptions backend with plans, seat counting, and usage accounting
    • NEWDelete organization or workspace directly from settings
    • NEWOrg domain auto-join — teammates with matching emails skip the invite step
    • IMPDashboard layout rebuilt — cleaner sidebar and components
    • IMPLanding page refreshed with updated hero, pricing, and nav
    • IMPTrace span viewer now renders content with rich formatting
    • IMPConfigurable per-plan rate limits replace one-size-fits-all throttling
    • IMPTraces now show which API key sent them
    • IMPFeedback popover simplified and streamlined
    • PRFTraces API refactored with root span detection and faster queries
    • FIXTime filter now persists correctly across page reloads
    • FIXDetail panel and monitoring table navigation edge cases resolved
    • DOCSDK docs overhauled — new quickstart and integration guides

    +12 under-the-hood changes shipped

  5. v0.1.0

    Observability goes live

    • NEWExecution traces with full context and filtering
    • NEWConversations view groups related traces together
    • NEWReal-time monitoring dashboard with timeseries charts
    • NEWPer-agent performance metrics and success rates
    • NEWCost tracking per trace, agent, and model
    • NEWEnvironment filtering for prod, staging, and dev
    • NEWPython SDK with @trace and @tool_call decorators
    • NEWOpenAI and Anthropic SDK adapters with streaming support
    • NEWAnnotate traces and spans with labels via `st.annotate()`
    • NEWManual span API — `start_span()` for custom instrumentation
    • NEW`user_id` propagates through spans and conversations
    • NEWAnthropic streaming captures thinking tokens alongside completions
    • NEWAPI key management and team settings
    • NEWNotification bell in the dashboard for invites and account events
    • NEWUsage page shows plan quotas and per-resource progress
    • IMPOrganization and workspace setup flow

    +45 under-the-hood changes shipped

  6. v0.0.1

    Landing page goes live

    • NEWFirst public landing page with hero, features, and waitlist
    • NEWWaitlist form live — sign up to be notified at launch
    • NEWSEO foundation — sitemap, robots.txt, and rich metadata
    • IMPDynamic OpenGraph and Twitter preview images
    • IMPMobile-responsive navigation with hamburger menu

    +18 under-the-hood changes shipped