Table of Contents

Tools

JD.AI provides 19 categories of built-in tools that the AI agent invokes automatically during conversations. You don't call tools directly — describe what you want in natural language and JD.AI selects the right tool. Each tool call is confirmed before execution unless you've enabled auto-run.

Tool execution example in the terminal

For full parameter documentation, see the Tools Reference.

Tool categories at a glance

Category Tools What it does
File read_file, write_file, edit_file, list_directory Read, write, and browse files
Search grep, glob Find text in files and locate files by pattern
Shell run_command Execute shell commands
Git git_status, git_diff, git_log, git_commit, git_push, git_pull, git_branch, git_checkout, git_stash Full git workflow
Web web_fetch Fetch a URL and return readable text
Web Search web_search Search the web for current information
Memory memory_store, memory_search, memory_forget Semantic memory for cross-turn recall
Subagent spawn_agent, spawn_team, query_team_context Delegate work to specialized agents
Think think Scratchpad for reasoning — no side effects
Environment get_environment System info: OS, runtime, disk space, tooling
Tasks create_task, list_tasks, update_task, complete_task, export_tasks Track work items
Code Execution execute_code Run snippets in C#, Python, Node.js, Bash, or PowerShell
Clipboard read_clipboard, write_clipboard Read from and write to the system clipboard
Questions ask_questions Present structured questionnaires to you
Diff / Patch create_patch, apply_patch Create and apply unified diffs atomically
Batch Edit batch_edit_files Multi-file text replacements in one atomic operation
Usage get_usage, reset_usage Token counts and cost estimates
Encoding & Crypto encode_base64, decode_base64, encode_url, decode_url, decode_jwt, hash_compute, generate_guid Encoding, decoding, hashing, and cryptographic utilities
Tailscale tailscale_status, tailscale_machines, tailscale_configure, tailscale_runner_probe, tailscale_export Tailscale Tailnet discovery and remote orchestration

File tools

Read, create, and edit files. The agent reads files to understand your code and writes files to make changes.

> read the first 20 lines of Program.cs
⚡ Tool: read_file(path: "Program.cs", startLine: 1, endLine: 20)

edit_file replaces exactly one occurrence of a string — surgical, minimal edits rather than full rewrites.

Search tools

Find text across your codebase with grep (regex search) or locate files with glob (pattern matching).

> find all files that reference ILogger
⚡ Tool: grep(pattern: "ILogger", glob: "**/*.cs")

Shell tools

Execute any shell command. The agent uses this for builds, tests, linting, and anything else that needs the terminal.

> run the tests
⚡ Tool: run_command(command: "dotnet test", timeoutSeconds: 120)

Git tools

A full git workflow — status, diff, log, commit, push, pull, branch, checkout, and stash. The agent drafts commit messages based on your changes and stages everything before committing.

> commit my changes with a descriptive message
⚡ Tool: git_commit(message: "feat: add input validation to registration form")

Web tools

Fetch any URL and get back readable text, or search the web for current information.

> search for the latest .NET 9 breaking changes
⚡ Tool: web_search(query: ".NET 9 breaking changes", count: 5)

Memory tools

Store and recall facts across the conversation using semantic memory. Useful for remembering architectural decisions, API patterns, or any context you want to persist.

> remember that the API key is stored in Azure Key Vault
⚡ Tool: memory_store(text: "API key is stored in Azure Key Vault", category: "architecture")

Subagent tools

Delegate focused work to specialized subagents that run in their own context window. This keeps your main conversation clean while offloading research, code review, or command execution.

Agent type Best for
explore Codebase research — read-only, no context cost
task Running commands and reporting results
plan Creating structured implementation plans
review Code review with actionable feedback
general Complex multi-step work

Think tool

A scratchpad the agent uses to reason through problems before acting. No side effects — it simply records a thought and moves on. You'll see these in the tool call stream as the agent plans its approach.

Environment tools

Returns system information — OS, architecture, runtime version, disk space, and installed tooling. Helps the agent write platform-appropriate commands.

Task tools

Track work items within a session. The agent can create tasks, mark them in progress or done, and export the full list as JSON.

Code execution tools

Run code snippets directly — supports C#, Python, Node.js, Bash, and PowerShell. Useful for quick validation, testing a regex, or running a one-off calculation.

> test this regex in python
⚡ Tool: execute_code(language: "python", code: "import re; print(re.findall(r'\\d+', 'abc 123 def 456'))")

Clipboard tools

Read from and write to the system clipboard. Cross-platform: uses PowerShell/clip on Windows, pbcopy/pbpaste on macOS, and xclip/xsel on Linux.

Encoding & Crypto tools

Encode, decode, hash, and inspect common developer data formats without leaving JD.AI.

> encode this string to base64
⚡ Tool: encode_base64(text: "Hello, World!")

> decode this JWT
⚡ Tool: decode_jwt(token: "eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9...")

> generate a new GUID
⚡ Tool: generate_guid(count: 1)
Warning

decode_jwt inspects token contents only — it does not verify the signature. Do not rely on it for authentication decisions.

MD5 and SHA1 are cryptographically weak. Prefer SHA256 or SHA512 for any security-sensitive hashing.

Tailscale Integration tools

Discover and orchestrate machines on your Tailscale Tailnet. Requires Tailscale CLI (tailscale) installed and authenticated, or API credentials configured via tailscale_configure.

> list all machines on my tailnet
⚡ Tool: tailscale_machines(filter: "online")

> check if the build server has a runner available
⚡ Tool: tailscale_runner_probe(target: "build-agent-01")

> export the machine list for automation
⚡ Tool: tailscale_export()

For security guidance and credential configuration, see the Tools Reference.

Tool loadouts

A Tool Loadout is a curated bundle of tools configured for a specific purpose. Instead of exposing every available tool to the agent (which increases token usage and can reduce tool-selection accuracy), a loadout defines exactly which tools are active.

JD.AI ships five built-in loadouts:

Loadout Tools included Best for
minimal Filesystem, Shell, think Simple scripts or token-constrained models
developer Minimal + Git, GitHub, Search, Analysis, Memory Code writing and review
research Minimal + Search, Web, Memory, Multimodal Web research and document analysis
devops Minimal + Git, Network, Scheduling Infrastructure and deployment tasks
full All tool categories General-purpose work (default when no loadout is set)

Loadouts are primarily a developer and integration feature. Subagents use them automatically — for example, explore subagents receive the research loadout, and task subagents receive the minimal loadout.

For developers building integrations, see Tool Loadouts (developer guide).

Every tool belongs to a safety tier that controls confirmation behavior:

Tier Behavior Examples
Auto-approve Runs without confirmation read_file, grep, glob, git_status, think
Confirm once Asks once per session write_file, edit_file, git_commit, spawn_agent
Always confirm Asks every invocation run_command, web_search, execute_code

Read-only tools run silently. Write operations ask once then auto-approve for the rest of the session. Shell commands and web searches always ask because they have the broadest potential impact.

Overriding confirmations

Mechanism Scope Description
/autorun Session Toggle auto-approve for all tools
/permissions Session Disable all permission checks
--dangerously-skip-permissions Process CLI flag — no confirmations at all
Warning

Disabling confirmations means the agent can write files, run commands, and commit code without asking. Use these overrides only in trusted, automated environments.

See also

Edit this page