Skip to main content
GitAgent ships with a set of built-in tools available in every agent without any installation. Enable them in agent.yaml or reference them directly in skills.

​
Enabling built-in tools

# agent.yaml
tools:
  web_search: true
  file_ops: true
  code_exec: false      # disabled by default — opt in explicitly
  http: true
Or use them directly in a skill: 
from gitagent.tools import web_search, read_file

@skill(name="research", description="Search the web for information")
def research(query: str) -> str:
    results = web_search(query, num_results=5)
    return "\n".join(r["snippet"] for r in results)

​
Available tools

Search the web using a configured search provider. 
from gitagent.tools import web_search

results = web_search(
    query="latest AI research papers 2024",
    num_results=10,         # default: 5
    provider="serper"       # serper | serpapi | brave | duckduckgo
)
# Returns: list of {title, url, snippet}
Configure the provider in agent.yaml: 
tools:
  web_search:
    provider: serper
    api_key: ${SERPER_API_KEY}

​
fetch_url

Fetch the content of a URL and return it as text. 
from gitagent.tools import fetch_url

content = fetch_url(
    url="https://example.com/article",
    extract_text=True,      # strip HTML, return plain text
    max_chars=5000          # truncate long pages
)

​
read_file

Read a file from the local filesystem (relative to the agent directory). 
from gitagent.tools import read_file

content = read_file("data/report.txt")
By default, only files within the agent directory can be read. Override the allowed paths in agent.yaml: 
tools:
  file_ops:
    allowed_paths:
      - ./data/
      - /tmp/agent-workspace/

​
write_file

Write content to a file. 
from gitagent.tools import write_file

write_file("output/summary.md", content="# Summary\n\n...")

​
list_files

List files in a directory. 
from gitagent.tools import list_files

files = list_files("data/", pattern="*.csv")

​
run_code

Execute a Python code snippet in a sandboxed environment. The sandbox has no network access and no filesystem access outside the agent’s workspace. 
from gitagent.tools import run_code

result = run_code("""
import json
data = [1, 2, 3, 4, 5]
print(json.dumps({"mean": sum(data) / len(data)}))
""")
# result.stdout, result.stderr, result.exit_code
run_code is disabled by default. Enable it explicitly in agent.yaml with tools.code_exec: true. Only enable it for agents where you fully control the input.

​
http_request

Make an HTTP request to any URL. 
from gitagent.tools import http_request

response = http_request(
    method="POST",
    url="https://api.example.com/data",
    headers={"Authorization": "Bearer token"},
    json={"key": "value"}
)
# response.status_code, response.json(), response.text

​
memory_read / memory_write

Read from and write to the agent’s persistent memory. These tools wrap the memory provider configured in agent.yaml. 
from gitagent.tools import memory_read, memory_write

# Write a fact
memory_write("user_preference", "prefers bullet-point summaries")

# Read it back in a future run
pref = memory_read("user_preference")

​
call_agent

Invoke another GitAgent as a sub-agent. 
from gitagent.tools import call_agent

result = call_agent(
    agent_path="./agents/researcher",   # path to another agent
    prompt="Research quantum computing advances in 2024"
)
The sub-agent runs with its own context but can share memory if configured.

​
Configuring tool defaults

# agent.yaml
tools:
  web_search:
    provider: serper
    api_key: ${SERPER_API_KEY}
    num_results: 10

  fetch_url:
    max_chars: 8000
    extract_text: true
    user_agent: "GitAgent/0.8"

  http:
    timeout: 30
    max_redirects: 5
    allowed_domains:          # restrict to specific domains
      - api.github.com
      - api.openai.com

  code_exec:
    timeout: 10               # max execution time in seconds
    memory_mb: 128            # max memory