ByteRover Knowledge Management

Use the brv CLI to manage your project's long-term memory.
Install: npm install -g byterover-cli
Knowledge is stored in .brv/context-tree/ as human-readable Markdown files.

No authentication needed. brv query, brv curate, and brv vc (local version control) work out of the box. Login is only required for remote sync (brv vc push/brv vc pull).

Workflow

1. 1. Before Thinking: Run brv query to understand existing patterns.
2. After Implementing: Run brv curate to save new patterns/decisions.

Commands

1. Query Knowledge

Overview: Retrieve relevant context from your project's knowledge base. Uses a configured LLM provider to synthesize answers from .brv/context-tree/ content.

Use this skill when:

• - The user wants you to recall something
• Your context does not contain information you need
• You need to recall your capabilities or past actions
• Before performing any action, to check for relevant rules, criteria, or preferences

Do NOT use this skill when:

• - The information is already present in your current context
• The query is about general knowledge, not stored memory

CODEBLOCK0

2. Search Context Tree

Overview: Retrieve a ranked list of matching files from .brv/context-tree/ via pure BM25 lookup. Unlike brv query, this does NOT call an LLM — no synthesis, no token cost, no provider setup needed. Returns structured results with paths, scores, and excerpts.

Use this skill when:

• - You need file paths to read rather than a synthesized answer
• You want fast, cheap retrieval with no LLM overhead
• You're in an automated pipeline that consumes structured results

Do NOT use this skill when:

• - You need a natural-language answer synthesized from multiple files — use brv query instead
• The information is already present in your current context

CODEBLOCK1

Flags: --limit N (1-50, default 10), --scope "domain/" (path prefix filter), --format json (structured output for automation).

3. Curate Context

Overview: Analyze and save knowledge to the local knowledge base. Uses a configured LLM provider to categorize and structure the context you provide.

Use this skill when:

• - The user wants you to remember something
• The user intentionally curates memory or knowledge
• There are meaningful memories from user interactions that should be persisted
• There are important facts about what you do, what you know, or what decisions and actions you have taken

Do NOT use this skill when:

• - The information is already stored and unchanged
• The information is transient or only relevant to the current task, or just general knowledge

CODEBLOCK2

Include source files (max 5, project-scoped only):

CODEBLOCK3

View curate history: to check past curations

• - Show recent entries (last 10)

brv curate view

• - Full detail for a specific entry: all files and operations performed (logId is printed by brv curate on completion, e.g. cur-1739700001000)

brv curate view cur-1739700001000

• - List entries with file operations visible (no logId needed)

brv curate view detail

• - Filter by time and status

brv curate view --since 1h --status completed

• - For all filter options

CODEBLOCK8

4. Review Pending Changes

Overview: After a curate operation, some changes may require human review before being applied. Use brv review to list, approve, or reject pending operations.

Use this when:

• - A curate operation reports pending reviews (shown in curate output)
• The user wants to check, approve, or reject pending changes

Do NOT use this skill when:

• - There are no pending reviews (check with brv review pending first)

Commands:

List all pending reviews for the current project:
CODEBLOCK9

Sample output:
CODEBLOCK10

Each pending task shows: operation type (ADD/UPDATE/DELETE/MERGE/UPSERT), file path, reason, and before/after summaries. High-impact operations are flagged.

Approve all operations for a task (applies the changes):
CODEBLOCK11

Reject all operations for a task (discards pending changes; restores backup for UPDATE/DELETE operations):
CODEBLOCK12

Approve or reject specific files within a task:

brv review approve <taskId> --file <path> --file <path>
brv review reject <taskId> --file <path>

File paths are relative to context tree (as shown in brv review pending output).

Note: Always ask the user before approving or rejecting critical changes.

JSON output (useful for agent-driven workflows):
CODEBLOCK14

5. LLM Provider Setup

brv query and brv curate require a configured LLM provider. Connect the default ByteRover provider (no API key needed):

CODEBLOCK15

To use a different provider (e.g., OpenAI, Anthropic, Google), list available options and connect with your own API key:

CODEBLOCK16

6. Project Locations

Overview: List registered projects and their context tree paths. Returns project metadata including initialization status and active state. Use -f json for machine-readable output.

Use this when:

• - You need to find a project's context tree path
• You need to check which projects are registered
• You need to verify if a project is initialized

Do NOT use this when:

• - You already know the project path from your current context
• You need project content rather than metadata — use brv query instead

CODEBLOCK17

JSON fields: projectPath, contextTreePath, isCurrent, isActive, isInitialized.

7. Version Control

Overview: brv vc provides git-based version control for your context tree. It uses standard git semantics — branching, committing, merging, history, and conflict resolution — all working locally with no authentication required. Remote sync with a team is optional. The legacy brv push, brv pull, and brv space commands are deprecated — use brv vc push, brv vc pull, and brv vc clone/brv vc remote add instead.

Use this when:

• - The user wants to track, commit, or inspect changes to the knowledge base
• The user wants to branch, merge, or undo knowledge changes
• The user wants to sync knowledge with a team (push/pull)
• The user wants to connect to or clone a team space
• The user asks about knowledge history or diffs

Do NOT use this when:

• - The user wants to query or curate knowledge — use brv query/brv curate instead
• The user wants to review pending curate operations — use brv review instead
• Version control is not initialized and the user didn't ask to set it up

Commands:

Available commands: init, status, add, commit, reset, log, branch, checkout, merge, config, clone, remote, fetch, push, pull.

First-Time Setup

Setup — local (no auth needed):
CODEBLOCK18

Setup — clone a team space (requires brv login):
CODEBLOCK19

Setup — connect existing project to a remote (requires brv login):
CODEBLOCK20

Local Workflow

Check status:
CODEBLOCK21

Stage and commit:
CODEBLOCK22

View history:
CODEBLOCK23

Unstage or undo:
CODEBLOCK24

Branch Management

CODEBLOCK25

Merge:
CODEBLOCK26

Set upstream tracking:
CODEBLOCK27

Cloud Sync (Remote Operations)

Requires ByteRover authentication (brv login) and a configured remote.

Manage remotes:
CODEBLOCK28

Fetch, pull, and push:
CODEBLOCK29

Clone a space:
CODEBLOCK30

8. Swarm Query

Overview: Search across all active memory providers simultaneously — ByteRover context tree, Obsidian vault, Local Markdown folders, GBrain, and Memory Wiki. Results are fused via Reciprocal Rank Fusion (RRF) and ranked by provider weight and relevance. No LLM call — pure algorithmic search.

Use this skill when:

• - You need to search across multiple knowledge sources at once
• The user has configured multiple memory providers (check with brv swarm status)
• You want results from Obsidian notes, GBrain entities, or wiki pages alongside ByteRover context

Do NOT use this skill when:

• - The user only has ByteRover configured — use brv query instead (it synthesizes via LLM)
• You need an LLM-synthesized answer — brv swarm query returns raw search results, not synthesized text

CODEBLOCK31

Output:
CODEBLOCK32

With explain mode (shows classification, provider selection, enrichment):
CODEBLOCK33

Output:
CODEBLOCK34

JSON output:
CODEBLOCK35

Output:
CODEBLOCK36

Limit results:
CODEBLOCK37

Flags: --explain (show routing details), --format json (structured output), -n <value> (max results).

9. Swarm Curate

Overview: Store knowledge in the best available external memory provider. ByteRover automatically classifies the content type and routes accordingly: entities (people, orgs) go to GBrain, notes (meeting notes, TODOs) go to Local Markdown, general content goes to the first writable provider. Falls back to ByteRover context tree if no external providers are available.

Use this skill when:

• - You want to store knowledge in an external provider (GBrain, Local Markdown, Memory Wiki)
• The user has configured writable swarm providers

Do NOT use this skill when:

• - You want to store in ByteRover's context tree specifically — use brv curate instead
• No swarm providers are configured — use brv curate instead

CODEBLOCK38

Output:
CODEBLOCK39

Target a specific provider:
CODEBLOCK40

Output:
CODEBLOCK41

CODEBLOCK42

Output:
CODEBLOCK43

JSON output:
CODEBLOCK44

Output:
CODEBLOCK45

Flags: --provider <id> (target specific provider), --format json (structured output).

10. Swarm Status

Overview: Check provider health and write targets before running swarm query or curate. Use this to verify which providers are available and operational.

Use this skill when:

• - Before running brv swarm query or brv swarm curate to check available providers
• Diagnosing why swarm results are missing from a specific provider

CODEBLOCK46

Output:
CODEBLOCK47

Data Handling

Storage: All knowledge is stored as Markdown files in .brv/context-tree/ within the project directory. Files are human-readable and version-controllable.

File access: The -f flag on brv curate reads files from the current project directory only. Paths outside the project root are rejected. Maximum 5 files per command, text and document formats only.

LLM usage: brv query and brv curate send context to a configured LLM provider for processing. The LLM sees the query or curate text and any included file contents. No data is sent to ByteRover servers unless you explicitly run brv vc push.

Cloud sync: brv vc push and brv vc pull require authentication (brv login) and sync knowledge with ByteRover's cloud service via git. All other commands operate without ByteRover authentication.

Error Handling

User Action Required: You MUST show this troubleshooting guide to users when errors occur.

"Not authenticated" | Run brv login --help for more details.
"No provider connected" | Run brv providers connect byterover (free, no key needed).
"Connection failed" / "Instance crashed" | User should kill brv process.
"Token has expired" / "Token is invalid" | Run brv login again to re-authenticate.
"Billing error" / "Rate limit exceeded" | User should check account credits or wait before retrying.

Agent-Fixable Errors:
You MUST handle these errors gracefully and retry the command after fixing.

"Missing required argument(s)." | Run brv <command> --help to see usage instructions.
"Maximum 5 files allowed" | Reduce to 5 or fewer -f flags per curate.
"File does not exist" | Verify path with ls, use relative paths from project root.
"File type not supported" | Only text, image, PDF, and office files are supported.

Quick Diagnosis

Run brv status to check authentication, project, and provider state.