Derived from karakeep-app/karakeep
issues + pull requests as shabda intake (external word, not verified fact).
Each cited item shows its verified author (resolved from the GitHub API, never inferred)
and weight. Inferred dependencies are marked as hypotheses.
Telos: The system becomes shippable as the 0.33.0 release — every issue and PR the maintainers assigned to the 0.33.0 milestone is resolved and the release can cut.
| Ref | Title | Verified author | Assoc. | Weight |
|---|---|---|---|---|
| #2669 | Removing large accounts locks the database (needs background reaping) | MohamedBassem | COLLABORATOR | high |
| #2172 | Uncertain origin of alpine-chrome image (supply-chain) | MemQu | NONE | high |
| #2798 | Autosaving tags can autoclose the edit window (loses edits) | NotaInutilis | NONE | high |
| #2594 | Collaborator links invisible in public lists | karabays | NONE | high |
| #2817 | Screenshot previewer lacks scroll bar | Summon528 | CONTRIBUTOR | high |
CITED After #110 — becomes a queryable AI knowledge layer. Established: the 0.33.0 milestone includes the AI-layer feature issues #441/#863. Treat as a real dependency.
HYPOTHESIS (kalpita) After #111 — stays bounded under heavy load. Not a maintainer-stated dependency. Inferred because shipping with known memory/CPU regressions is risky — validate with the maintainers before treating it as a hard gate.
Not on a committed milestone. The AI layer (#110) and resource bounding (#111) feed the release above; faithful capture (#112) and client reliability (#113) are independent quality backlogs that can run in parallel afterward.
Telos: Bookmarks become a queryable knowledge base — semantic search and chat over saved content, with agents able to fetch and reason over bookmark content, not just keyword lookup.
| Ref | Title | Verified author | Assoc. | Weight | State |
|---|---|---|---|---|---|
| #2857 | Bookmark embedding support (semantic-search foundation) | MohamedBassem | COLLABORATOR | high | MERGED |
| #2861 | AI assistant component prep | MohamedBassem | COLLABORATOR | high | MERGED |
| #2820 | Agentic search content + bookmark slicing | vbalko-claimate | CONTRIBUTOR | high | OPEN |
| #2684 | Automated tagging prompt/model evals | MohamedBassem | COLLABORATOR | high | OPEN |
| #2785 | Configurable LLM temperature (default 0) | MohamedBassem | COLLABORATOR | high | OPEN |
| #2890 | Native Anthropic (Claude) inference provider | johnford2002 | NONE | medium | OPEN |
Unblocks: the 0.33.0 release (CITED dependency).
Telos: Large imports, crawls and deletions run within bounded memory and CPU and cannot corrupt data or exhaust the host — steady-state resource use stays stable over time.
| Ref | Title | Verified author | Assoc. | Weight | State |
|---|---|---|---|---|---|
| #2625 | Split Docker builders + trim worker image | kapdon | NONE | low | OPEN |
| Ref | Title | Verified author | Assoc. | Weight |
|---|---|---|---|---|
| #1748 | High memory usage on worker during large imports | Garrett-R | NONE | high |
| #2344 | Memory leak in 0.30 (RAM grows over time) | nkkfs | NONE | medium |
| #2519 | Parser OOM leaves orphaned assets / invalid data | BryanWall | NONE | medium |
| #2369 | Chromium stuck with high idle CPU after large imports | Emstar | NONE | medium |
| #2397 | Docker CPU usage higher since 0.29.3 | raidolo | NONE | medium |
Unblocks: the 0.33.0 release — but only under the hypothesised dependency; confirm with the maintainers.
Telos: Saved pages, feeds and embeds reproduce their source faithfully — archives render, metadata is honored, partial-asset failures don't fail the page, and blocked sources have a sanctioned crawl path.
| Ref | Title | Verified author | Assoc. | Weight | State |
|---|---|---|---|---|---|
| #2804 | Screenshot precrawled archives | MohamedBassem | COLLABORATOR | high | OPEN |
| #2626 | Custom metascraper-twitter plugin | kapdon | NONE | medium | OPEN |
| #2783 | Render Distill math in reader view | Mykidzou | NONE | low | OPEN |
| Ref | Title | Verified author | Assoc. | Weight |
|---|---|---|---|---|
| #1977 | Full-page archives no longer render after CSP blocklist | dvshenry | NONE | high |
| #1667 | Image-download failure fails the whole crawl | Fmstrat | NONE | high |
| #2885 | Reddit crawling now getting blocked | MohamedBassem | COLLABORATOR | high |
| #2304 | SingleFile support broken in 0.90.2/0.90.3 | mapleshadow | NONE | medium |
| #2334 | Reddit RSS feed ignores provided metadata (date/author) | leonbeon | NONE | medium |
| #2205 | Youtube videos show incorrect author (hosted) | Rab-C | NONE | medium |
Unblocks: standalone reliability quality; no downstream anantara dependency.
Telos: Karakeep's client surfaces behave correctly and predictably on every platform — the Android app no longer crashes on specific content, the web UI orders lists consistently, and the server binds to the network interfaces it is configured for.
| Ref | Title | Verified author | Assoc. | Weight |
|---|---|---|---|---|
| #1581 | List sorting inconsistent in dropdown | enkelmedia | NONE | high |
| #2596 | Specific bookmark crashes Android app | Djuuu | NONE | medium |
| #2404 | Cannot bind listen address to all interfaces | robled | NONE | medium |
Unblocks: standalone client reliability; no downstream anantara dependency.
The Backlog milestone (GitHub milestone #5 — "Things that the project maintainers will want to work on it at some point", 12 open / 7 closed) is the maintainers' deferred pool, not committed near-term work. Several of its issues already surface above as drivers of the four quality directions (e.g. #1748, #1667, #1977, #1581, #1953). Nothing in Backlog is scheduled for the 0.33.0 cut.
Source of truth: the repo's open issues + open/recently-merged PRs, landed in an NKS graph as shabda (epistemic kalpita). Directions are the bianhua (qualitative transformations) discerned over that field.
Weight combines authorAssociation (OWNER/MEMBER/COLLABORATOR/CONTRIBUTOR = strong signal; NONE = a user need, kept when substantive) with engagement and milestone membership. Milestone membership outranks reaction count.
Dependencies: CITED = grounded in a maintainer milestone/statement; HYPOTHESIS (kalpita) = the maintainer's own architectural inference, not established fact. The one inferred edge (#111 → #109) is hedged accordingly.
Authors are verified, never inferred — a community contribution is never attributed to the maintainer.