AgentFlight
Local-first flight recorder for AI coding agents.
- Platform
- Node.js 20+, macOS, Linux, Windows
- Pricing
- Free CLI, open-core
Overview
- Built
- 2026
- Version
- 0.3.1
- Category
- Developer Tools / AI coding workflow
- Platform
- Node.js 20+ (macOS, Linux, Windows)
- Distribution
- npm / npx
- CLI
- agentflight
- License
- Apache-2.0
AgentFlight sits around tools like Codex, Claude Code, Cursor, Windsurf, Gemini CLI, Aider, and OpenCode. It records what happened during an AI coding session, captures proof, highlights risk, renders a replay timeline, and writes a handoff prompt for the next agent or reviewer.
AI coding agents move fast, but you still need to know what changed, what passed, what failed, what looks risky, and how to safely continue the work later. AgentFlight records that context locally. It does not call an LLM, upload source code, or replace your coding agent. It gives you and the next agent a proof trail.
AgentFlight is the control room around your agent, not another agent. ProjScan reads your codebase structure, AgentLoopKit holds the work to a task contract, and TokenTrace tracks what your AI CLIs cost. AgentFlight is the flight recorder: it proves what a session did and what to do next.
Features
What it does well.
Session recording
`agentflight start` opens a local session with task title, branch, commit, dirty state, package manager, and detected tool availability, so every later artifact has context.
Verification evidence
`agentflight verify -- <command>` runs proof commands and records the command, timestamps, duration, exit code, pass or fail status, and the stdout and stderr paths.
Changed-file risk
`agentflight status` groups changed files by rough risk area and explains the next safest action.
Snapshots
`agentflight snapshot --note "..."` records a checkpoint with git state, changed files, risk level, verification summary, and a human note.
Proof reports
`agentflight report` writes a local Markdown report with task, timeline, changed files, risk, verification evidence, recommendation, and next action.
Replay timelines
`agentflight replay` writes a self-contained HTML replay showing session events, snapshots, verification cards, changed file groups, and readiness.
Resume prompts
`agentflight resume` creates a Codex or Claude-ready prompt with current state, proof gaps, snapshot context, next action, and guardrails.
Doctor checks
`agentflight doctor` checks Node, npm, git, package manager, config, writable paths, the current session, ProjScan, AgentLoopKit, and common scripts.
ProjScan and AgentLoopKit adapters
AgentFlight integrates defensively with ProjScan for repo intelligence and AgentLoopKit for task discipline and verification. It fails gracefully when either tool is unavailable.
Local-first privacy
AgentFlight stores human-readable files under `.agentflight/`. Sessions, reports, evidence, and current state stay local and are git-ignored by default.
See it in action
A closer look.
The session as a timeline
`agentflight replay` writes a self-contained HTML file showing session start, verification attempts, snapshots, report generation, and the resume prompt. Open it in a browser, with no server and no upload.
Closer look
Three local artifacts.
AgentFlight writes everything to `.agentflight/`. The replay, the proof, and the handoff are files you can open, diff, and share.
- The session as a timeline
- `agentflight replay` renders session start, verification attempts, snapshots, report generation, and resume generation as one local HTML artifact you open in a browser.
- Proof without pretending
- Passed commands, failed commands, exit codes, stdout, stderr, and durations are stored as local evidence. AgentFlight does not claim tests passed unless the command actually passed.
- The next agent gets the state
- The resume prompt includes the task, changed files, risk, latest snapshot, verification state, proof gaps, next action, and guardrails that keep the work scoped.
Trust
Local files. No telemetry. No cloud upload.
AgentFlight reads git status and package metadata, runs verification commands only when you ask, and writes local artifacts under `.agentflight/`. Reports include filenames and summaries by default, not full code diffs.
- Nothing leaves the machine
- No telemetry, no login, no cloud sync, and no source upload. AgentFlight does not call an LLM and has no account to create.
- What AgentFlight is not
- Not a coding agent, not a cloud service, not a replacement for tests, not a security scanner, not a CI platform, and not a code review replacement.
- Records, never overwrites your work
- AgentFlight captures context and writes its own artifacts under `.agentflight/`. It runs your verification commands only when you invoke `verify`.
- Where it sits in the studio
- AgentFlight is the user-facing flight recorder and proof layer. ProjScan is the repo intelligence engine, AgentLoopKit is the task discipline and verification workflow engine, and TokenTrace is local AI CLI usage analytics.
Start, verify, snapshot, replay, resume.
One workflow runs from the first agent prompt to a reviewable handoff. Run your coding agent between start and verify. Local-first, Apache-2.0, no cloud.
Initialize in a repo
Start a session
Capture verification
Snapshot a checkpoint
Check changed files and risk
Write a proof report
Render the replay timeline
Write a resume prompt
More from the studio
ProjScan
Agent-first code intelligence and Mission Control for developer agents. ProjScan turns a plain-language goal into a mission, a local proof loop, a review gate, and a resumable handoff, all over MCP with no source upload.
View
AgentLoopKit
AgentLoopKit is a local-first acceptance layer for coding agents. It adds task contracts, verification evidence, review-readiness scoring, PR preparation, run history, maintainer checks, and reviewer handoffs to existing repos.
View
TokenTrace
Reads what Claude Code, Codex, and other AI CLIs write on disk. Shows cost, models, projects, and sessions in a local dashboard. No cloud, no telemetry.
View