Skip to content

BUILD WITH YOUR AI CODING TOOL

You make the engineering decisions.
Shifu handles the context.

Get started
shifuPut your context to work
Coding agent
Build exports
Shifu / Context engineRelevant context retrieved
FROM YOUR WORK

Existing customers still use the legacy export fallback.

↳ Design discussion + incident notes
PLAN

Add CSV exports. Preserve the fallback and check both paths.

Illustrative workflow

Connect where decisions happen.

  • Google Meet
  • Coding harnesses
  • Work files
  • Slack Coming soon

Use context where you already work.

  • OpenCode
  • Claude Code
  • Cursor
  • Codex
  • Grok Bot Coming soon

More agents to follow.

Your context. Your connections.

Account-scoped access. Revoke connections to stop future sync.

Privacy & data handling

Raw coding prompts and session transcripts stay local. Work-file uploads are not locally redacted. Connected content uses hosted AI processing; retained records require a deletion request.

CONTEXT, EVALUATED

Better answers.
With your work behind them.

Five engineering tasks. Same model and inputs, with and without Shifu. Engineering plan quality scored by Jev out of 10.

Internal pilot · One work history · One answer per condition · Three controls excluded.

Without Shifu5.01/10
With Shifu7.46/10

+2.44 points
GPT-6.1 Sol
Same model · low reasoning
Engineering plan quality

Engineering scenarios
Without ShifuWith Shifu
Retry recovery
Without Shifu: 3.36 out of 10
With Shifu: 5.46 out of 10
Identity isolation
Without Shifu: 5.14 out of 10
With Shifu: 6.33 out of 10
Empty-result handling
Without Shifu: 6.08 out of 10
With Shifu: 8.04 out of 10
Correction persistence
Without Shifu: 4.84 out of 10
With Shifu: 8.47 out of 10
Historical backfill
Without Shifu: 5.66 out of 10
With Shifu: 8.99 out of 10

Model-rated answer quality, not measured correctness or customer impact.

Eval methodology & limitations

Five history-dependent engineering tasks were selected from real work before answer generation. Three controls are reported separately and excluded from the headline. Both conditions used fresh isolated Codex sessions, GPT-6.1 Sol at low reasoning, identical task inputs and one answer per condition. The with-Shifu condition also received the current integration skill and read-only production context retrieval. Project-specific expected constraints and source references were held out from the answering models, except in the full-context control. Explicit records from earlier benchmark activities were filtered from retrieved context; the filter is heuristic. Jev 1.13.0 scored each answer on five anchored dimensions, normalized to 0–10. Each answer score is the equal-weight mean of those dimensions; the headline gives each of the five primary tasks equal weight. Condition labels were hidden from the judge. Both answers were judged using the same task-specific reference packet, retrieved evidence and integration instructions. Those materials verify assertions; expected requirements and current user instructions remain authoritative.

This internal pilot covers one work history and five selected primary tasks, with one answer per condition. It does not establish statistical significance or represent all engineers, repositories or tasks.

Scores estimate the quality of pre-coding engineering plans. Generated implementations and tests were not executed; productivity, production safety and customer outcomes were not measured.

The with-Shifu condition combines integration instructions and retrieval, so the result does not isolate memory alone. Production retrieval is bounded and the live corpus was not frozen.

Reference documents can contain local, uncommitted contracts or proposals. Their presence in production memory and their deployed behavior were not independently established. A retrieval gap may reflect ingestion or selection.

The full-context control repeats one primary scenario and is not an independent task domain. Setup gains may reflect instructions supplied by the integration skill.

Scenario descriptions are anonymized paraphrases, not verbatim benchmark prompts. Private prompts, generated answers, source records, account identifiers and implementation details are not published. Scores are unchanged.

Evaluated 24 Oct 2026. Judge: Jev 1.13.0.

Read the detailed engineering report ↗Download aggregate results & methods ↗

Try it on your
next feature.

Start with one supported source and a feature you’re planning. Inspect the context, apply what fits, and add what’s missing.

Get startedMake your work compound. Free during beta.