Agentic TerminalAn AI agent that ships from your terminal.
Describe the change you want. The agent reads your codebase, plans the work, edits the files, and runs your tests. You review the diff and decide what stays.
- Build and plan agents
- Any model provider
- LSP-aware
- MCP servers
- Git-native
- Runs your tests
Every session starts with the whole repo in view.
The agent attaches to your workspace with your model, your project context, and a safe approval policy already set, so the first thing you type can be the task itself.
Agent session· ~/app
Running- Model
- Sonnet 5Switch provider any time
- Context
- Full repoFiles, git history and LSP
- Agent
- buildTab switches to plan
- Approvals
- Ask firstBefore edits and commands
- Status
- RunningWorking in ~/app
- Session
- 42 minResumable at any time
Defaults shown. Models, approval rules and agents can be changed per project or per session.
An engineer in the terminal, not a chat window
The agent works where your code already lives. It uses your tools and your tests, and it stops to ask you before anything important.
Agentic by default
Give it a goal instead of a prompt. It breaks the task into steps and works through them until it is done.
Understands your repo
It reads files, searches symbols and follows imports with language-server awareness, so its changes match your patterns.
Plan before it builds
A read-only plan agent explores and proposes. Hand the plan to the build agent when you agree with it.
Runs your toolchain
It installs dependencies, runs builds and tests, and reads the failures. MCP servers add any other tool it needs.
You stay in control
Edits and shell commands wait for your approval unless you allow them. Every change can be undone.
Bring any model
Use frontier models from any major provider, or a local model, and switch between them in the middle of a session.
One command to start. Plain language after that.
The agent is a single command in any repository. Everything else is a sentence describing what you want done.
Start in any repository
Run the agent from the root of a project and it opens with that repository as its context. For a one-off job, use agent run to hand over a single task. Use --continue to pick up the last conversation where you left it.
# Open any repository and start the agentcd ~/appagent# Or hand it one task and let it run to the endagent run "explain how requests reach the database"# Pick up where you left off yesterdayagent --continueWhat can you hand to the agent?
Most tasks that take you an afternoon of reading, editing and re-running tests can be described in one sentence.
Build features
Starts with
$agent run "add CSV export to reports"Describe the behavior you want. The agent finds where it belongs, writes it, and wires it up end to end.
Fix bugs
Starts with
$agent run "fix the failing auth test"Point it at a failing test or an error message. It reproduces the problem, finds the cause, and proves the fix.
Understand a codebase
Starts with
$agent run --plan "how does billing work?"Ask how something works in plan mode and get an answer that cites the actual files, without any edits.
Refactor safely
Starts with
$agent run "move date helpers to lib/date"Rename, move and restructure code across many files, with the test suite run after each step.
Write tests
Starts with
$agent run "cover utils/price.ts with tests"Cover untested modules with tests that follow your existing conventions and pass on the first run.
Review changes
Starts with
$agent run --plan "review my git diff"Have the agent read your diff before you push. It flags bugs, missed edge cases and inconsistencies.
From request to reviewed change in four steps.
There is no new editor to learn and no copying code out of a chat window. The agent works in your repository, and you keep the final decision.
- Select a step to pin it.
Autonomy with guardrails
The agent can change code and run commands, so it is limited the same way you would limit a new teammate. It needs permission to act, its changes can be reversed, and there is a record of what it did.
Permission prompts
edit · bash: askFile edits and shell commands ask first by default. You can allow specific commands per project, such as your test runner, and keep everything else gated.
Read-only planning
plan agentThe plan agent cannot edit files. Use it on unfamiliar or sensitive code to understand the work before anything changes.
Reversible changes
agent undoEvery run is recorded as one change set. agent undo restores the files exactly as they were before the run.
Isolated workspaces
containerIn the cloud, the agent runs inside your own workspace container. It can reach only the projects and resources that you can reach.
Your models, your keys
BYOKUse your organization's provider accounts or a local model. Provider keys are stored as secrets and are never written into the repository.
Full transcript
transcriptEvery prompt, tool call and command is saved in the session history, so you can see exactly how a change was made.
Matched rule
Reading the repository is allowed
Tool calls in this session
Select a call to see how it was handled.
Put an agent in your terminal.
Security & compliance
flowbit runs on independently audited security, privacy and cloud controls — so your apps and data meet the standards your customers expect.
