Agent Tools
The agent has access to built-in tools organized by category. Tool availability depends on the current mode.
Planning Tools
| Tool | Description | Plan | Execute |
|---|---|---|---|
ask_user | Ask structured questions (2-4 options + free text per question) | yes | no |
plan_tool | Create/update task plans, track resources, generate completion reports. In Execute mode, create and add operations require user approval. | yes | yes |
Research Tools
| Tool | Description | Plan | Execute |
|---|---|---|---|
papers | Search papers, read ArXiv, get citations, find code/datasets | yes | yes |
web_search | Brave web search | yes | yes |
research | Spawn an independent research sub-agent with its own context | no | yes |
github_read_file | Read files from GitHub repos | yes | yes |
github_list_repos | List repos for a user/org | yes | yes |
github_find_examples | Find code examples by filename pattern | yes | yes |
github_search_repos | Search GitHub repositories by query | yes | yes |
github_get_readme | Get README from a repository | yes | yes |
Papers Operations
| Operation | Source | Description |
|---|---|---|
search | OpenAlex | Full-text search with year/citation filters (primary) |
arxiv_search | arXiv | Search arXiv directly — best for ML/CS/Physics preprints |
semantic_search | Semantic Scholar | Search with abstracts, good for recent papers |
trending | OpenAlex | Recent highly-cited papers |
details | OpenAlex + CrossRef | Full metadata, abstract, OA links |
read_paper | arXiv (ar5iv) | Parse HTML into sections, read by name/number |
citations | OpenAlex | References + cited-by with batch fetching |
recommend | OpenAlex | Related works |
author_papers | OpenAlex | Papers by a specific author |
find_code | Papers With Code | GitHub repos linked to papers |
find_datasets | Papers With Code | Datasets linked to papers |
Search source recommendations
- arXiv search: Latest ML/AI preprints, CS research, physics papers
- OpenAlex search: Broad academic coverage across all fields, great for citation data
- Semantic Scholar: When you need abstracts or Semantic Scholar IDs
Research Sub-Agent
The research tool spawns an independent sub-agent with its own context window. The parent agent sees nested tool calls streamed in real-time. Useful for deep dives that would consume too much of the main conversation's context.
Writing Tool
| Tool | Description | Plan | Execute |
|---|---|---|---|
writing | Paper authoring — manage outline, write sections, update bibliography | no | yes |
The writing tool manages a writing project stored in the database:
- Outline: Define paper structure (sections, subsections)
- Sections: Write/update individual sections with auto-save
- Bibliography: Manage citations and references
- Auto-save: All changes persist to the database immediately, surviving across workers and restarts
Paper preview and client-side export (Markdown/LaTeX) are available in the Paper tab in the UI.
Filesystem Tools
| Tool | Description | Plan | Execute |
|---|---|---|---|
read | Read files with line numbers | yes | yes |
write | Create/overwrite files | no | yes |
edit | Find-and-replace in files | no | yes |
inspect_files | Parallel file reading with relevance filtering | yes | yes |
list_dir | List directory contents | yes | yes |
glob_files | Find files by glob pattern | yes | yes |
grep_search | Search file contents | yes | yes |
In Plan mode, only read-only filesystem tools are available.
inspect_files
The inspect_files tool reads multiple files or entire directories concurrently and scores each file for relevance against a query. Useful when the agent needs to scan a directory to find which files are relevant without reading them one by one.
- Expands directories to file listings (hidden files excluded)
- Reads all files in parallel via
asyncio.gather - Scores relevance by keyword overlap with the query
- Returns relevant files with content, skips low-relevance files
- Safety limits: 50 files max, 200 lines per file, 2MB file-size gate
Execution Tools
| Tool | Description | Plan | Execute |
|---|---|---|---|
bash | Execute shell commands (Docker-isolated when available) | no | yes |
sandbox | Run code in Docker containers, SSH remotes, or Modal cloud | no | yes |
Sandbox Types
| Type | Description |
|---|---|
| Local (Docker) | Docker container on the host machine |
| SSH | Remote machine via SSH |
| Modal | Cloud sandbox via Modal |
Workspace Tools
| Tool | Description | Plan | Execute |
|---|---|---|---|
workspace | Project workspace operations — knowledge graph, notes, search, failure logs | yes | yes |
Workspace Operations
| Operation | Description |
|---|---|
status | View workspace summary (file counts, knowledge graph size, recent failures) |
search | Search files by name or content |
note | Save a research note with topic and content |
knowledge_add | Add entity to the knowledge graph |
knowledge_relate | Add relationship between entities |
knowledge_query | Search entities in the knowledge graph |
knowledge_summary | Get full knowledge graph context for the conversation |
recent_failures | View recent tool/API failure logs |
See Projects & Workspaces for details on the knowledge graph entity and relationship types.
Mode Restrictions
Tools are filtered based on the current mode before being sent to the LLM. See Modes for details on the enforcement layers.
In summary:
- Plan mode:
ask_user,plan_tool,workspace, read-only filesystem,inspect_files, web search, papers, GitHub, HuggingFace, compute planning, and MCP tools configured for Plan mode - Execute mode: Everything except
ask_user, plus MCP tools configured for Execute mode - MCP tools: Filtered by the per-server mode configuration set in Settings > MCP Servers (default: both modes)