Local AI coding with a shared Brain and Git sync
How local provider CLIs, hosted coordination data, encrypted transcript chunks, and Git shadow refs work together in Amoeba.
Claude Code and Codex run as local processes under each teammate’s provider login. Model inference still happens through that provider, and prompts plus required code context follow the provider account’s data controls.
Amoeba’s hosted Brain stores workspace and session records, plans, locks, memories, comments, activity, and encrypted transcript chunks after built-in pattern-based secret redaction. Amoeba encrypts transcript chunks and stored GitHub tokens before they are written to the database, and other workspace records rely on our database provider’s storage encryption. Amoeba holds the keys, so this is not end-to-end encryption. Prompt-derived context can also be processed by Anthropic for overlap checks and task summaries.
Repository content remains in local checkouts and synchronizes through Git shadow refs pushed to the configured remote under the user’s credentials. The Brain stores snapshot commit and ref metadata, not a repository copy.
Transcript chunks follow the organization’s configured transcript-retention period. Other coordination records do not share that automatic expiry. Provider-local and provider-hosted histories follow their own controls.