Agent features
Built for real local work
PTY
Terminal access
The agent can run commands in a real terminal when write mode allows it.
FS
File tools
It can read, write and patch files inside a chosen workspace.
LOG
Audit trail
Destructive actions are controlled and logged so you can review what happened.
Tasks
What to ask a local AI agent to do
- Fix a failing test and explain the changed files.
- Add a small feature while preserving the existing project style.
- Inspect an unfamiliar repo and summarize the architecture.
- Create a script, run it, then repair errors until it works.
- Refactor a narrow module with tests or command-line verification.
Example agent prompt
Goal: add a /health endpoint and a test.
Scope: inspect the server routes and test folder first.
Constraints: keep the change minimal and preserve existing style.
Verification: run the project test command.
Deliverable: summarize changed files and test output.
Safety
Local control matters
Agent prompts often include private source code, file paths, logs, environment details and product plans. Saient keeps the agent local, with write mode off by default and visible control over destructive actions.
The agent is powerful, so keep tasks scoped. Ask it to inspect first, plan second, edit third and verify last.
Give the agent a goal and let it verify the work.
Saient combines a local AI agent with chat, image generation, video generation, vision, TTS and LoRA tools.