Description
infaiw is a daemon-first agent harness built for your terminal, enabling efficient delegation of tasks to AI agents while maintaining human oversight. It operates as a background service, ensuring that work continues even if the terminal is closed. The core design principles revolve around durable, branching session timelines, sidecar loops with acceptance checks, and explicit human approvals for critical actions.
Read More
One of infaiw's standout features is its sidecar loop mechanism. A sidecar is a delegated subagent whose completion is only recognized after a parent-defined bash script exits successfully within a specified turn budget (20, 40, or 100 turns). This prevents agents from prematurely declaring victory and ensures their output meets predefined criteria. The acceptance script's output is fed back into the sidecar's conversation if it fails, guiding the agent to correct its work. This system provides checked answers, meaning the parent session receives findings that have already passed a verification step, rather than mere promises.
The daemon-first architecture allows sessions to persist and communicate via inter-session mailboxes (ISAC), routed by the engine. Users can attach to sessions from multiple clients, leave, and return to find their work in progress. A central screen lists all active sessions, their types, models, working directories, and status, with sidecars nested under their parents for clear visibility into their progress and budget usage. This design ensures low memory footprint, with the entire daemon and its sessions consuming under 100 MiB, making it efficient for long-running tasks.
infaiw is engineered for long-running work, offering features like branching timelines that allow users to revert to any past event and explore alternative paths without losing previous attempts. Approvals are granular, with tools categorized as `allow`, `human`, or `deny`. Write, edit, or bash scripts pause for explicit human approval, even if the client window is closed. Users can define and save named policies for different sessions. Transparency into token usage is provided, showing context fullness, last turn's cost, and the proportion of prompts from cache. Sessions can be exported as comprehensive HTML reports, including all sidecars, branches, and tool calls, for easy sharing and record-keeping. The tool also supports switching models mid-session, allowing users to leverage different models like Claude Code, OpenAI Codex, or DeepSeek for various stages of a task, all while maintaining the same timeline and approvals. MCP (Multi-Agent Communication Protocol) servers can be added and managed directly from the TUI or shell, with OAuth authorization and policy integration. The underlying storage mechanism is git-like, using append-only timelines and content-addressed blobs for efficient, instant loading and scrolling of even thousands of events. It also supports multimodal input, allowing users to paste screenshots for model processing and receive generated images directly in the chat.
- CategoryAI Agent Builders
- Pricing ModelFree
- DeploymentLocal
- PlatformsmacOS, Linux
infaiw's Core Features
Daemon-first agent harness for terminal
Durable, branching session timelines
Sidecar loops with acceptance checks
Human-in-the-loop approvals for consequential steps
Inter-session messaging via ISAC engine
Low memory footprint (under 100 MiB)
Centralized session management and monitoring
Configurable turn budgets for sidecars
Bash-script based acceptance checks for model-agnostic verification
Ability to branch sessions from any point in history
Granular approval policies (allow, human, deny) for tools
Transparency into token usage and context fullness
Export full session timelines as filterable HTML reports
Switch models mid-session (e.g., Claude Code, OpenAI Codex, DeepSeek)
MCP server support with OAuth and policy integration
Git-like append-only timeline storage for efficient history
How to use infaiw?
Install infaiw: Install the infaiw package using Homebrew for macOS/Linux or your distribution's package manager.
Start the service: Initiate the infaiw user service to run as a background daemon.
Attach to a session: Open the TUI client and attach to an existing session or create a new one.
Delegate tasks: Hand off tasks to sidecar agents, defining a bash acceptance script and a turn budget.
Approve actions: Review and approve or deny consequential actions proposed by agents, even if the client is closed.
Monitor progress: Use the TUI to view all running sessions, their status, and sidecar activities.
Branch timelines: If an agent takes a wrong turn, branch the session from an earlier point on the timeline.
Export reports: Generate an HTML report of the entire session history for review or sharing.
infaiw's Use Cases
- Automated Code Review
- Complex Software Development
- Billing Integration Review
- Technical Documentation Generation
- System Configuration Management
- Data Analysis and Reporting
- Multi-Model Workflow Orchestration
- Agentic Pipeline Development
- Collaborative Agent Work
- Incident Response Automation



