Integration

Scan Pinecone for memory poisoning, prompt injection and leaked secrets

Pinecone is a common managed home for agent memory: each vector carries metadata with the remembered text, often split by namespace per user or tenant. Mimvo lists and fetches the vectors in one namespace, checks the text and the vectors, and reports which records to review, quarantine or delete.

Install#

Mimvo is an open-source Python package for Python 3.11+, published on PyPI. The pinecone extra adds the Pinecone client and nothing else.

bash
pip install "mimvo[pinecone]"

Scan Pinecone from the command line#

One command reads every record, runs the secrets, injection and poisoning checks, and writes a self-contained HTML report you can forward.

bash
mimvo scan pinecone \  --index agent-memory \  --namespace prod \  --text-field content \  --report report.html
mimvo scan pinecone options
OptionWhat it does
--indexIndex name.
--namespaceNamespace to scan. Default is the default namespace.
--api-keyAPI key. Defaults to PINECONE_API_KEY; use a read-scoped key.
--hostIndex host, for serverless indexes.
--text-fieldMetadata field that holds the text.
  • Use a read-scoped API key. --api-key defaults to the PINECONE_API_KEY environment variable.
  • Scan one namespace at a time with --namespace; without it, Mimvo reads the default namespace.
  • Set --text-field to the metadata field that holds the text, and --host for a serverless index.

Scan from Python#

The same scan from your own code, for a scheduled job or a notebook.

python
from pathlib import Pathfrom mimvo import Mimvofrom mimvo.scan import PineconeScanSource, render_html# Reads PINECONE_API_KEY from the environment.source = PineconeScanSource(index="agent-memory", namespace="prod", text_field="content")report = Mimvo().scan(source)report.source = "pinecone:agent-memory"Path("report.html").write_text(render_html(report), encoding="utf-8")

What Mimvo finds in Pinecone#

Thirteen finding codes across three checks. Each finding comes with a severity, a recommended action, the evidence (secrets masked), fix steps, and an OWASP and CWE mapping. Details are in the finding code reference.

Mimvo finding codes
FindingCodeSeverityAction
Leaked secretsecret_detectedCriticaldelete
Personal datapii_detectedMediumreview
Hidden instructionpersistent_instructionHighreview
Canary instruction droppedknown_answerMediumreview
Injection embeddingembedding_injectionMediumreview
Poisoned factmemory_poisoningHighquarantine
Destination redirectdestination_redirectHighreview
Adversarial textadversarial_textMediumreview
Tampered vectorembedding_mismatchMediumquarantine
Contradicts older memorytemporal_contradictionMediumreview
Answer flipretrieval_flipMediumreview
Poisoning clusterpoisoning_clusterLowreview
Retrieval hubhub_recordLowreview

Gate CI on the results#

Fail the job on high-severity findings, send results to GitHub code scanning as SARIF, and put a summary on the run page. See Use in CI for a full GitHub Actions workflow.

bash
mimvo scan pinecone \  --index agent-memory \  --namespace prod \  --text-field content \  --fail-on high \  --sarif mimvo.sarif \  --markdown "$GITHUB_STEP_SUMMARY"

Is it safe to run against production Pinecone?#

Lists and fetches vectors with metadata. Never upserts or deletes.

The default detectors run on your machine with no API key and no model download, so memory text does not leave it. Secret values are masked in every report. Hosted detectors are opt-in.

Try it without connecting anything

Open the live dashboard to browse a sample scan, or read the guide to AI agent memory poisoning.

Other integrations#