Bring Your Own Keys (BYOK)
Run Agentic Foundation agent work on your Cursor or Anthropic account instead of the platform’s managed Cursor key.
What BYOK is
Bring Your Own Keys (BYOK) is the named tier for workspaces that supply their own Agent Execution credentials:
- Cursor User API Key (default for new workspaces), or
- Claude API key (Claude Code CLI)
With BYOK selected, Agentic Foundation scans and automated Fix it jobs bill and authenticate to your Cursor or Anthropic account. The platform does not use its managed Cursor execution key for those runs.
BYOK is not “no subprocessors.” It reduces ThinkAI-controlled AI execution for agent runs. Platform hosting and your chosen AI vendor still process data as described below.
Public Trust Center detail: Security Overview and Subprocessors.
How to enable it
- Open Settings → Integrations → Agent Execution Provider.
- Choose Cursor User API Key or Claude.
- Paste a valid execution key (not an Insights admin key), Test, then Save.
- Leave ThinkAI Managed Cursor Agent selected only if you intend to use the platform’s managed Cursor account.
During onboarding, the same picker appears under Integrations setup.
| Option in the app | Tier | Whose key runs the agent |
|---|---|---|
| Cursor User API Key | BYOK | Your Cursor user / service-account key |
| Claude | BYOK | Your Anthropic API key (sk-ant-api…) |
| ThinkAI Managed Cursor Agent | Managed | Platform Cursor execution account |
Execution keys are separate from AI Insights admin keys (seat/spend analytics). You can connect Insights without changing which key runs Agentic Foundation. See Cursor and Claude.
What still leaves your environment under BYOK
Even with BYOK, the following still leave (or are processed outside) your corporate network when you use the product:
- Chosen AI vendor for agent runs. Prompts and repository excerpts used for readiness grading, fix generation, and (when Code Chat reuses Agent Execution) chat/wiki synthesis go to Cursor or Anthropic under your key and their terms. BYOK changes whose account is charged, not whether an AI vendor sees that content.
- Platform infrastructure. Workspace data at rest and in transit is hosted by the platform’s infrastructure providers (today: cloud hosting and Postgres, authentication, operational logging, transactional email). See the infrastructure list on Subprocessors.
- Source control and other integrations you connect. GitHub (and any other tools you authorize) still receive the data those integrations need, for example repos, webhooks, and fix PRs.
- Insights admin APIs (if connected). Cursor or Claude admin keys still pull usage and spend aggregates into the platform. That path is independent of BYOK execution.
Agentic Foundation workers check out selected repos into ephemeral job environments on platform compute, run the vendor CLI, then persist scores and issues in the workspace database. Missing BYOK credentials do not silently switch to the managed Cursor key for readiness; the run fails or completes only the filesystem scan until you fix the key.
What BYOK does not include
Do not treat BYOK as:
- Customer-managed encryption keys (CMEK) for platform disks or databases
- Dedicated per-tenant compute nodes or “air-gapped” isolation from all vendors
- Removal of platform infrastructure subprocessors
- A guarantee that no AI vendor ever sees repository excerpts
Those stronger isolation options are separate commercial topics, not this tier.