← VantraChangelog
What we've shipped, in order.
May 7, 2026
Agent loop detection, simple evals, p95 latency, rate limit visibility, cost forecast
- • Agent loop alert: fires when any single trace exceeds a span count threshold
- • Eval score alert: fires when thumbs-up rate on recent traces drops below threshold
- • Thumbs up/down rating on trace detail page — rating indicator shown in trace list
May 7, 2026
Dashboard: p95 latency, rate limits, cost forecast, truncation detection
- • Overview now shows p95 latency (with avg as secondary) instead of avg alone
- • Rate-limited spans (429s) surfaced in the error rate card
- • Monthly cost projection shown in the cost card, based on 7-day trailing average
- • Truncated response detection in trace waterfall: flags LLM spans where output hit the token limit
May 7, 2026
Pricing update
- • New Indie plan: $9/mo, 500K spans/month, 60-day retention
- • Pro: $19/mo, 2M spans/month (was $29, 1M)
- • Team: $79/mo (was $299)
- • Not per seat, ever
- • Docs FAQ now searchable with accordion UI
May 6, 2026
v0.1.3 — Streaming, prompt versioning, trace search, weekly digest
- • Streaming support:
stream=True calls now captured automatically for both OpenAI and Anthropic, in Python and TypeScript - • Prompt version tagging: pass
prompt_version="v2" to @vantra.trace and filter traces by version in the dashboard - • Version comparison: select two prompt versions to see error rate, latency, and cost side by side
- • Trace search: filter by name, status, date range, or prompt version
- • Daily budget alerts: get notified when spend exceeds a threshold, resets at midnight UTC
- • Weekly digest: Monday morning email with traces, tokens, cost, error rate, and top models per org
- • Python SDK: 141 unit tests + 53 real-API integration tests
- • TypeScript SDK: 31 unit tests + 18 real-API integration tests
May 6, 2026
v0.1.2 — Async support, capture_io, full message history
- • Python SDK:
@vantra.trace now works on async def functions - • Python SDK:
capture_io=False suppresses prompt/response capture for sensitive workloads - • Python SDK:
span.set_input() for manually logging span inputs - • Python SDK: fixed Anthropic integration — tokens and cost now tracked correctly
- • Python SDK: full conversation history captured, not just the last message
- • TypeScript SDK: fixed async trace timing — duration and tokens now accurate
- • TypeScript SDK: Anthropic patching fixed for
@anthropic-ai/sdk
May 4, 2026
v0.1.1 — TypeScript SDK, Slack alerts, team invites
- • TypeScript / Node.js SDK published to npm (
npm install vantra-sdk) - • Slack webhook alerts — get notified in any channel
- • Team invites — add teammates to your org by email
- • Anomaly alerts for error rate, p95 latency, and cost per hour
May 1, 2026
v0.1.0 — Launch
- • Python SDK on PyPI —
pip install vantra - • Auto-patches OpenAI and Anthropic on
init() - • Waterfall trace view with per-span timing
- • Cost tracking per call, per model, per day
- • Dashboard with real-time traces and cost overview
- • Free tier: 50K spans/month, 30-day retention
Ship something worth mentioning? Follow on GitHub