Skip to content

BACKLOG.md

Latest commit

 

History

History
58 lines (44 loc) · 3.37 KB

BACKLOG.md

File metadata and controls

58 lines (44 loc) · 3.37 KB
Raw

Backlog

Sentro's forward roadmap — tracked publicly. Vote via GitHub issues. Last updated: 2026-04-14.

Shipped (v0.2.0)

  • Session grouping — group related runs into conversation threads (sessionId, userId on AgentRun, /api/projects/:id/sessions)
  • LLM proxy mode — zero-code instrumentation via /api/v1/proxy/chat/completions (OpenAI) and /api/v1/proxy/messages (Anthropic)
  • Prompt management — Prompt/PromptVersion models, CRUD API, sentro.getPrompt() / sentro.get_prompt() in SDKs
  • Evals / scoring — Score model, sentro.score(runId, name, value) in both SDKs; supports human, LLM-as-judge, programmatic
  • Drift and guardrail alerts — auto-detect looping agents, token burn, repeated tool calls; fires drift_detected webhook
  • Datasets — save runs as test fixtures, sentro.getDataset(name) / sentro.get_dataset(name); import runs as items
  • Session replay UI — animated timeline scrubber on run detail page (play/pause, 1x-10x speed, click-to-seek)
  • Playground — edit and re-run any LLM call from the UI (OpenAI-compatible and Anthropic providers)
  • Dashboard pages for sessions, prompts, scores, datasets — full UI coverage of v0.2.0 backend features
  • Dataset run evaluations — sentro.runEval() / sentro.run_eval() with built-in evaluators (exactMatch, contains, regexMatch)
  • Demo/seed data script — npm run seed:demo populates a dashboard with realistic runs, errors, prompts, datasets, and scores

Shipped (v0.1.1)

  • TypeScript SDK on npm
  • Python SDK on PyPI
  • Code coverage: TS 96%, Python 94%
  • Event webhooks (5 event types, HMAC signing, filters)
  • Security hardening: SSRF protection, rate limiting, session cleanup
  • GitHub Actions CI pipeline
  • CORS middleware for cross-origin SDKs
  • Framework integrations: Claude Code, OpenClaw, LangChain, CrewAI, Vercel AI SDK
  • One-line installers for Claude Code and OpenClaw
  • OTLP ingestion — accept OpenTelemetry traces at /api/v1/traces

Up Next (ordered by bang-for-buck)

Tier 2 — Differentiating

  • Protobuf OTLP — /api/v1/traces now decodes application/x-protobuf alongside JSON
  • Batch eval dashboard — see eval results over time ✅ done (daily trend chart on Scores page); compare versions still TODO
  • Human annotation queue — UI for human raters to score runs in bulk

Tier 3 — Scale & enterprise

  • Source maps — deobfuscate minified stack traces
  • Redis — caching + BullMQ job queue
  • ClickHouse — event analytics at scale
  • SaaS mode — multi-tenancy, teams, RBAC
  • SSO / SAML
  • Audit log
  • Grafana dashboard template

Tier 4 — Nice to have

  • [~] Helm chart / Railway / Render / Fly.io one-click deploy — README now documents the paths for Railway/Fly/Render on top of the existing Dockerfile. Proper Helm chart and verified platform templates still TODO.
  • Screenshots in README — present in docs/screenshots/, referenced in the Dashboard section. Note: pre-v0.3 sidebar refactor; refresh when time permits.
  • OTLP gRPC endpoint

Contributing

Pick anything from this list and open a PR. For larger items, open an issue first to align on approach.