Pricing
[code]smith is priced at $1.75 per agent compute unit. Agent compute units are a composite metric that combines LLM time, sandbox resources and context window size into one metric. Today, agent compute units map to approximately 1 minute of LLM thinking time from the latest frontier models. Subagents accrue agent compute units independently. Autofix is billed based on the agent compute units [code]smith uses to fix your PR. We’ve intentionally priced [code]smith to support frontier levels of intelligence. As [code]smith develops, we expect to offer dedicated intelligence tiers for different tasks, and will look to price those competitively.Cost controls
[code]smith automatically suspends a session when it first reaches $100 in cumulative spend. To resume the session, you must confirm that you wish to continue via the web UI for that session.Progressive billing
[code]smith bills incrementally for usage. The first time your combined monthly usage including runners crosses $100, you will be invoiced. If a charge fails, access to [code]smith may be restricted. CI runners will continue to run and accrue charges. You can restore access by paying your invoice from Billing and Usage.Context usage
Agent compute units scale with context usage. Long-running sessions will have to maintain more context and compact context frequently, which can drive up costs significantly. Prefer starting new sessions when doing new work.Data retention and sharing
We use a mix of LLM model providers, all configured with zero data retention. Some models, including the latest Anthropic models, require data retention. Admins can explicitly opt-in to these models from [code]smith onboarding or the Settings page. Prompts and repository context are used only to complete your task. We do not train on user code. To enable [code]smith to make cross-repo changes, we create an index based on the codebases you store with us. You can disable this index from Settings. For broader privacy terms, see the privacy policy.Connector permissions
When you connect MCP tools, each server has an Availability scope in [code]smith:My Session (Personal)
Only your [code]smith sessions can use the connector. Credentials are stored for your user account and are not shared with teammates. Note that other users may still view output and sessions that use this server.
Team
Every org member’s [code]smith sessions can use the server. The org stores one shared OAuth or API credential.