Direction, not commitments. What works today lives in the README and docs.gettidings.com; what shipped when lives in
CHANGELOG.md. Priorities shift with community feedback — if something here matters to you, open an issue and say so.
Tidings is built by one maintainer plus contributors: an email-first, self-hosted finance journal. No Plaid, no bank APIs. Five Canadian banks ship today; the architecture is country-neutral and parsers for institutions anywhere are a contribution surface.
Rough order of interest, not a schedule:
- More parsers — the highest-leverage contribution. Email alerts for new institutions, and PDF statement coverage beyond RBC and Simplii. See the parser tutorial.
- Notification provider contributor guide — a "how to add Pushover / Discord / Slack" worked example, mirroring the parser tutorial.
- Unified
deploy.shfor the AWS path — one idempotent orchestrator withinstall/update/teardown/statussubcommands. - Shareable demo deep links — URLs into the hosted demo with preloaded state.
- Native MCP entry-point — a
finance-mcpstdio server wrapping the bearer-token API, so MCP-native agents (Claude Desktop and similar) connect without hand-writing HTTP tools. - Release tooling — adopt release-please once the manual release ritual feels like overhead (around release 5–10), and pin GitHub Actions to commit SHAs before 1.0 so two builds of the same tag can't differ.
- Conversational assistant — "ask an agent about your transactions" over the existing API. Speculative; only if it's genuinely useful rather than gimmicky.
- No Plaid or bank-API integration. Email-first, no-credential ingestion is the identity of the project.
- No multi-user / family mode. Tidings is a single-household journal.
- No native mobile app. The responsive web UI is the mobile story.
- No zero-based budgeting methodology. Tidings shows where money went; it doesn't assign every dollar before you spend it.
Open an issue or GitHub Discussion — "I want X" or "X matters more than Y" is useful signal. PRs are welcome on anything above, parsers especially; comment on an existing issue before starting so work doesn't double up.