[FR] Improve AI search, avoid hitting maximum context length

Offen

#776 erstellt am 22.1.2026 2 Kommentare 1 Reaktion 0 zugewiesene Personen Auf GitHub ansehen

Dieses Issue hat noch niemand übernommen.

Bewertung

Schwierigkeit: 5/5
Geschätzter Aufwand: Über eine Woche
Anfängerfreundlichkeit: 25/100

Issue-Typ: Feature
Klarheit: Muss geklärt werden
Aktivitätsstatus: Veraltet
Tech-Stack: typescript
Bereich: ai, search

Rechercherichtung

No files, tests, or implementation entry points are named. Start by locating the Ask/global search implementation and how it builds model context, then determine how repository selection and context-limit failures are handled. Done would require a defined strategy that lets global searches answer file-location questions without exceeding the model context limit.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Beschreibung

ask_sb Feature

I think the AI search ("Ask") feature could be improved a lot. Right now, you can quickly select one repository or do a global search with one click. It's quite tedious to select individual repos and if you have a global question it's often unclear which repositories you should select in the first place. Sourcebot' most powerful feature is to be able to search globally.

nfortunately, it's too easy to hit the maximum context size of a model. I think the AI feature should be a bit more intelligent on what it does. For example, one of the most frequent questions we have is: "Where do I find...."? For example: "Where do I find the arrow icon image file?" etc.

In this case, it should perhaps first find out which files might be affected at all (PNG, JPEG, SVG). Then it should search over filenames and paths only and see whether it could answer the question that way. Perhaps it should even automatically try to segment the search and search iteratively over chunks of e.g. 16 repositories. I think a lot more stuff could be done to make this feature work better. Right now most global searches end without an answer due to context limits.

Vorherrschende Sprache: TypeScript
Sterne: 3.9k
Forks: 374
Ø Merge: 21 Std. 18 Min.
Gemergte PRs (30 T.): 39

Beitragsleitfaden

Beitragsleitfaden öffnen

Erste Schritte

Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
Forken Sie das Repository und arbeiten Sie in einem Branch.
Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.

Mehr aus sourcebot-dev/sourcebot

[bug] Search result hydration performs duplicate repository lookups within the same chunk Offen

bug triage needed

Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 84/100

sourcebot-dev/sourcebot#1681 · 1 Kommentar · Erstellt am 22.9.2026
Linux ctags build instructions in CONTRIBUTING.md are missing build dependencies Offen

Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 84/100

sourcebot-dev/sourcebot#1633 · Erstellt am 05.9.2026
Generic Git URL configs keep percent-encoded repo names Offen

bug

Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 78/100

sourcebot-dev/sourcebot#1384 · 4 Kommentare · Erstellt am 29.6.2026
MCP ask_codebase rejects explicit languageModel: getLanguageModelKey includes displayName which the MCP schema doesn't expose Offen

ask_sb bug

Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 76/100

sourcebot-dev/sourcebot#1137 · 4 Kommentare · Erstellt am 20.4.2026
[bug] GitHub App installation tokens (ghs_) are rejected by the credential preflight Offen

bug triage needed

Schwierigkeit 3/5 1-2 Tage Anfängerfreundlichkeit 48/100

sourcebot-dev/sourcebot#1661 · 1 Kommentar · Erstellt am 16.9.2026

Alle Issues in sourcebot-dev/sourcebot