Skip to content

ROADMAP.md

Latest commit

 

History

History
76 lines (52 loc) · 4.06 KB

ROADMAP.md

File metadata and controls

76 lines (52 loc) · 4.06 KB
Raw

Roadmap

Nitpick is moving toward a source-agnostic, reviewable incident-workflow layer: bring errors in from the systems a team already uses, investigate them with the agent backend it prefers, and send trustworthy findings to the places where work already happens.

The roadmap is directional rather than a promise of dates. Adoption feedback, security, and operational reliability can change the order. Completed work is kept visible so the project’s progress is easy to assess.

Product direction

More log sources → one normalized incident model → more agent backends
                                      ↓
                         docs, issues, notifications, fix PRs
                                      ↓
                    local-first today → deployable tomorrow

The design principles stay constant as the surface area grows:

  • Reviewable automation — investigations explain their evidence; changes arrive as PRs.
  • Provider choice — sources, agent backends, and sinks should be replaceable adapters.
  • Operational control — budgets, permissions, retention, and approval gates remain visible.
  • Deploy anywhere — start locally, then support dependable self-hosted operation.

Shipped in v0.1.0

  • Datadog log ingestion with fingerprinting and SQLite deduplication.
  • Claude CLI investigation with cost limits and structured RCA output.
  • Markdown and Notion RCA sinks; GitHub and Linear issue trackers; console and Slack notifications.
  • Optional fix PRs in isolated git worktrees, disabled by default.
  • Local dashboard, credential-free demo mode, tests, CI, release packaging, and contributor/security guidance.

Now

The next focus is making the existing local workflow dependable enough for real teams:

  • Expand provider contract tests, cache migration coverage, and sanitized log fixtures.
  • Improve dashboard accessibility, filtering, run history, and error-to-investigation navigation.
  • Make privacy, retention, permissions, and cost behavior configurable and easy to audit.
  • Improve setup and scheduling documentation for macOS, Linux, and repeatable upgrades.

Next

More sources

  • Add Sentry, OpenTelemetry/OTLP, and CloudWatch adapters behind the existing source interface.
  • Support webhook and fixture inputs so teams can integrate systems without a bespoke polling loop.
  • Preserve one normalized error model so downstream investigations do not depend on the source vendor.

More sinks and workflow integrations

  • Add Jira and additional issue/project-management adapters.
  • Add richer incident destinations such as PagerDuty-style notifications and generic webhooks.
  • Improve document output with linked timelines, evidence sections, and reusable RCA templates.

More agent backends

  • Add first-class Codex support alongside Claude CLI.
  • Introduce a provider-neutral agent runner with explicit capability, permission, and cost contracts.
  • Make model selection and investigation prompts configurable without coupling the pipeline to one vendor.

Deployability

  • Provide a documented Docker/Compose path for the pipeline and dashboard.
  • Add production-oriented systemd and launchd examples with health checks and log rotation.
  • Make database, logs, configuration, and secrets paths explicit so Nitpick can run cleanly in a container or worker host.

Later

  • Add richer incident timelines, related-error grouping, and cross-service correlation.
  • Add approval gates before issue creation, external notifications, or fix-PR creation.
  • Support a GitHub App integration so teams can minimize local CLI coupling and scope permissions precisely.
  • Evaluate a hosted or multi-user deployment model only after self-hosted operation is stable and privacy boundaries are proven.

How priorities are chosen

We prioritize work that reduces the time from a real error to a reviewed, well-documented action. Issues with sanitized examples, measurable user impact, and a clear privacy or operational model are especially useful. Open an issue with the problem and desired outcome.