What We Receive and Store
RepoQL is operated by Clanker Gear Limited. This page explains the information we process and the controls available to you. The Privacy Policy explains purposes, recipients, retention and privacy rights more fully.
- We do not use your data to train AI models and require providers receiving it from us not to use it for model training.
- Cloud request payloads are processed for the request and then discarded. Separately retained parameters, error messages and feedback can contain code or other material supplied by you or your agent.
- Signed out, the software sends no repository content or tool-usage telemetry. It contacts the update service for manual updates or automatic update checks you enable; those requests expose normal connection information.
- Signed in, tool-usage telemetry is collected unless disabled. It is deleted from the primary store after 30 days, with all operational copies expiring within 90 days.
- Cached embeddings contain vectors and content-derived hashes rather than source text. They are treated as confidential and can be deleted on an authorised request.
Cloud processing
Cloud embeddings process repository text and context or search queries. Reranking processes queries and candidate text. Inference processes prompts and relevant context. Uplink relays authorised requests and results between clients and your RepoQL instances.
These request payloads are processed in memory, forwarded to the provider needed for the feature and discarded. They are not retained as payload records in application storage or logs. This statement does not cover the separate retained records below.
Voyage AI provides cloud embeddings and reranking; Fireworks AI provides inference. We use those services under zero-retention, no-training configurations. Google Cloud hosts our main cloud infrastructure. Other supporting providers and their roles are identified in the Privacy Policy.
Retained records
- Tool-usage telemetry: one record per signed-in tool call unless disabled. It includes supplied parameter values, truncated at 1,024 characters per value, and error messages. Parameters can contain queries, questions, SQL or scripts, including pasted code, credentials or personal information. The channel does not separately attach repository files, the index or returned tool results. Fields are listed below.
- Installation and account information: telemetry includes software, operating-system and agent information and an installation identifier. Our server associates records with the authenticated user and organisation.
- Cached embeddings: vectors and hashes used to avoid repeated computation. Cached results may serve matching requests from different customers using the same source identity and content-derived cache keys. Hashing does not establish anonymity. Cache confidentiality and deletion commitments apply, together with any stricter applicable DPA restrictions.
- Billing and service records: request counts, durations, token counts and repository identifiers used for billing and rate limits, together with account and subscription information.
- Uplink audits: organisation-visible metadata about members, instances, tools, targets, outcomes, timing, payload sizes and clients. These audits do not include relayed payload content.
- Feedback: the message you submit and host diagnostics concerning memory, performance and index statistics, without attaching file contents. Feedback text also appears in operational logs.
- Website analytics and replay: Datadog receives website and portal usage, performance, device and connection information. Session replay can capture page content and interactions. Portal analytics are associated with signed-in account and organisation information. These records are separate from binary telemetry.
The 30-day and 90-day limits apply to binary tool-usage telemetry, not to every category. The Privacy Policy describes retention for other records. Disabling telemetry or cancelling a subscription does not itself delete previously received records.
Telemetry controls
Set DO_NOT_TRACK=1 in the environment used to start the host, or set telemetry.enabled = false in RepoQL settings, to disable binary tool-usage telemetry. The environment is checked on host startup; settings changes take effect after restarting the host.
Set telemetry.params = false to remove parameter values from subsequent records after restart. Parameter names and other fields remain, including error messages that can echo supplied material.
These controls do not disable cloud processing you request, billing and service records, Uplink audits, update requests, feedback submissions or website analytics. The software's environment variable is not a browser tracking preference. Website choices are described in the Privacy Policy.
Tool-usage fields
| Field | Who fills it | What it is |
|---|---|---|
tool |
the binary | Which tool ran — explore, read, query, … |
door |
the binary | How it was called — mcp, grpc, or cli |
session |
the binary | Correlates calls within one agent conversation |
agent |
the binary | The calling agent's self-reported name and version |
outcome |
the binary | OK, ERROR, ZERO_RESULT, BUDGET_REFUSED, CANCELLED |
error_class, error_message |
the binary | The error you were shown |
elapsed_ms, tokens_used, result_count |
the binary | How long it took, how big the answer was, how much came back |
params |
the binary | The parameters you passed, verbatim — globs, SQL, search terms, questions, and sandbox source — truncated at 1,024 characters per value |
version, os, arch |
the binary | Build identity, operating system and version, process architecture |
machine_id |
the binary | A random identifier for this installation, stable across runs — not derived from your hardware |
user_id, org_id |
our server | Taken from your account token, never from anything the binary claims |
These records may also be sent to Datadog for monitoring, within the same 90-day operational-copy limit. Disabling binary telemetry stops new records being sent through that channel and to its monitoring destination; it does not delete older records or disable separate website analytics.
For access, correction, deletion or privacy enquiries, contact support@repoql.ai.