ByteRover CLI
A persistent, versioned project-memory layer for AI coding agents.
Per-dimension scores and reasoning
Evidence shows: CLI provides review workflow (approve/reject), indicating user confirmation; but least privilege, data flow transparency, sensitive data handling are not detailed. Many dependencies, but no security audit or vulnerability scan evidence. External effects include cloud sync, file operations, but not clearly limited. Rollback exists (vc reset), but data recovery guarantees not specified. Source attribution: publisher unverified, but author info in code. Deductions: least privilege, data flow transparency, sensitive data handling, dependency security, external effects, rollback, source attribution are only partial or not fully documented.
Evidence shows: code and tests consistent, tests cover critical paths, error messages structured (errors[]). But dependency availability unverified, and many dependencies may introduce risk. Deductions: dependency availability unverified, failure messages structured but not comprehensive.
Evidence shows: targeted at AI coding agents, scenarios clear (curate/query), capability boundaries documented (command list), trigger precision high (command args), environment fit (multi-platform, Node>=20). Deductions: capability boundaries not fully explicit (e.g., tool permissions), environment fit unverified.
Evidence shows: information architecture clear (detailed README), install notes clear (curl/npm), naming stable (brv), examples abundant, license clear (Elastic-2.0), but known limitations not explicit, version changelog not provided, maintenance responsibility not clear. Deductions: known limitations, version changelog, maintenance responsibility missing.
Evidence shows: output usable (--format json), marginal value high (provides memory layer), cost-benefit not assessed. Deductions: cost-benefit not assessed.
Evidence shows: README claims benchmark results, but no reproduction method; tests exist, but not independently verified; facts and inferences not clearly separated. Deductions: claim traceability insufficient, cross-source corroboration insufficient, fact-inference separation insufficient.
- Publisher identity unverified, proceed with caution.
- Many dependencies without security audit, supply chain risk.
- External effects like cloud sync and file operations not clearly limited, may affect user data.
- Benchmark results lack reproduction method, need independent verification.
What does this agent do, and when should you use it?
ByteRover CLI, invoked as brv, is an interactive REPL and project-knowledge memory tool for AI coding agents. It runs from a project directory, uses an agentic map to understand the codebase, and can read and write files, execute code, and store knowledge in a context tree. Developers can curate and query that context through the TUI, web dashboard, or command line, with a review workflow for pending changes. The context tree has Git-like branching, commits, merges, and push/pull operations, and can sync through ByteRover Cloud for team sharing. The repository documents 20 LLM providers, an MCP server, Hub and connector support, and compatibility with 22+ coding agents including Cursor, Claude Code, Windsurf, and Cline.
Running brv in a project starts an interactive REPL that auto-configures on first use. A developer can run /curate "Auth uses JWT with 24h expiry" @src/middleware/auth.ts to add a statement and file reference to knowledge storage, then run /query How is authentication implemented? to query the context tree. brv curate, brv query, and brv review pending/approve/reject handle knowledge curation, retrieval, and review of pending operations; brv vc init, add, commit, branch, merge, push, and pull version and synchronize the context tree. brv providers connect/switch and brv model switch configure model access, brv mcp starts the MCP server, and brv webui opens the dashboard. Configuration is persisted in settings.json under the global brv data directory and takes effect after brv restart.
- A developer using Claude Code, Cursor, or a similar coding tool who needs authentication rules, architecture decisions, and implementation context to persist across sessions.
- An engineer maintaining a large repository who wants to use brv curate to turn important files and project conventions into searchable context.
- A team that needs to share project knowledge through ByteRover Cloud push/pull while managing spaces, members, and permissions in the web dashboard.
- A developer exploring parallel implementations who needs vc commit, branch, checkout, and merge for the project context itself.
- A developer using local or hosted models who needs to choose and switch LLM providers through providers connect, providers switch, and model switch.
What are this agent's strengths and limitations?
- Stores project knowledge in a queryable context tree with explicit curate, query, and review operations instead of relying only on transient chat history.
- Provides separate Git-like version control for the context tree, including branches, commits, merges, remotes, and push/pull.
- Offers a React/Ink interactive TUI, web dashboard, JSON-formatted CLI output, and an MCP server for both interactive and automated workflows.
- Documents 20 LLM providers plus an OpenAI-Compatible option, reducing dependence on a single model service.
- Supports worktree links and read-only knowledge sources so subdirectories can share project context without creating nested .brv directories.
- The npm installation path requires Node.js >= 20; the no-Node bundled installer is documented only for macOS and Linux.
- Cloud collaboration requires ByteRover Cloud authentication or an API key, and team sync, spaces, and permission management depend on that hosted service.
- Model responsiveness and reliability depend on the selected provider; local models may require higher request and iteration timeout settings.
- Settings changes require brv restart because the daemon does not watch settings.json while it is running.
- License evidence is inconsistent: the supplied repository metadata says NOASSERTION, while the README states Elastic License 2.0.
How do you install or deploy this agent?
On macOS or Linux, run: curl -fsSL https://byterover.dev/install.sh | sh. This bundled path does not require Node.js and supports macOS ARM64, macOS x64, Linux x64, and Linux ARM64. On all platforms, with Node.js >= 20 installed, run: npm install -g byterover-cli. Verify with brv --version, then enter a project directory and run brv; first run auto-configures. For ByteRover Cloud, sign in through the dashboard or run brv login with a ByteRover API key.
How do you use this agent?
From the target project, run brv to start the REPL. Try /curate "Auth uses JWT with 24h expiry" @src/middleware/auth.ts, then /query How is authentication implemented? to confirm curation and retrieval. Run brv webui for the dashboard; commands support --format json for scripting. To version project context, run brv vc init, brv vc add, and brv vc commit; use brv vc push and brv vc pull for version-controlled cloud sync. Run brv providers list to inspect providers, then use brv providers connect and brv model switch to configure a model.
How does this agent compare with similar options?
ByteRover is positioned as a persistent project-context and knowledge-management layer for coding agents such as Cursor, Claude Code, Windsurf, and Cline, rather than as a documented replacement for those coding clients. Its stated distinction is the context tree, review workflow, and Git-like version control around that tree.