01
"Why did the AI write it this way?"
Intent, alternatives considered, and decision rationale stay paired with the commit. The "why" you can't read from the diff is one query away, even six months later.
Open betav2.0 · 4 CLI agents supported · solo to team
AI wrote the code. We capture the why.
AIFlare records every Claude Code, Codex, Gemini, and Copilot session alongside its git commit. Intent, alternatives considered, and decision rationale stay next to the code — a searchable history of how AI-written code actually came to be.
skill
curl -fsSL https://aiflare.dev/install.js -o install.cjs && node install.cjs && rm install.cjsOpen beta · Free to startNo card required
aiflare.dev / aiflare-core
live · 87ms
aiflare-core / Timeline
main · 248 entriesLast 7dAll agents
- Isolate env vars across parallel test runs9f3a8b1 · 2d ago · soph · Claude Code · 27 turnsBUGFIXtests
- Switch webhook retries to exponential backofff72e4d1 · 3d ago · dipree · Codex · 18 turnsFEATUREwebhooks
- Refactor Org · Team · Project permission checkse71a0b3 · 6d ago · soph · Claude Code · 41 turnsREFACTORauth
- Split session-summary insight pipelined59f1a4 · 1w ago · dipree · Codex · 32 turnsREFACTORinsights
- Context Inject — one-line sessionId injection flowc4d82f6 · 1w ago · dipree · Claude Code · 11 turnsFEATUREcontext-inject
CAPTURE CONTEXT
Same tools you already use, context just shows up.
AIFlare doesn't add a CLI or change your workflow. It runs as a Skill in Claude Code, Codex, Gemini, and Copilot, so as you work normally the intent, decisions, and rejected alternatives are captured for you.
Install the skill once
First, drop the aiflare.yml you downloaded from aiflare.dev into your project root. Then one command lets you pick which agents to install for (Claude Code, Codex, Gemini, Copilot — one or more), installs seven Skills, hooks, and the MCP server, and merges settings · MCP · CLAUDE.md / AGENTS.md for you.
my-project %Re-run
$ curl -fsSL https://aiflare.dev/install.js -o install.cjs && node install.cjs→ validating aiflare.yml ok→ selecting agents (Claude Code · Codex · Gemini · Copilot) ok→ installing 7 Skills · hooks · MCP server ok→ merging settings · MCP · CLAUDE.md / AGENTS.md ok
Ready. Capture runs automatically from your next git commit.$ ▍Claude Code, Codex, Gemini, Copilot — all four major CLI agents supported.
The same context-capture Skill runs across all four agents — Claude Code, OpenAI Codex CLI, Gemini CLI, and GitHub Copilot CLI. Support for Cursor and other agents is on the roadmap.
- Built on the context-capture Skill — an open Agent Skills standard
- Install once, no SDK code to write
- Captures prompts, decisions, and rejected alternatives
C
Claude Code
anthropic · v2.x+
Supported
O
OpenAI Codex CLI
openai · cli
Supported
G
Gemini CLI
google · gemini-2.5
Supported
◆
GitHub Copilot CLI
github · copilot
Supported
●
Cursor
cursor.sh · agent mode
Planned
Sessions flow with your commits
Every agent conversation is captured at the moment of git commit. Intent, alternatives, and the diff summary are stored as a pair with the commit — once pushed, your team sees the same context.
- capturedIsolate env vars across parallel test runs9f3a8b1 · 2d ago · soph · Claude Code
- capturedSwitch webhook retries to exponential backofff72e4d1 · 3d ago · dipree · Codex
- capturedRefactor Org · Team · Project permission checkse71a0b3 · 6d ago · soph · Claude Code
- capturedContext Inject — one-line sessionId injection flowc4d82f6 · 1w ago · dipree · Claude Code
Your entire history, at a glance
Search, filter, auto-summarize. The reason behind a decision from six months ago is one query away — and you can re-inject it into the next session with a single sessionId line.
- Full-text search · author, tag, and session filters
- Session summaries and prompt evaluation auto-generated
- Sharing through Org · Team · Project permissions
Session · 9f3a8b1
27 turns · 12 files · 4m 18sPROMPT · TURN 1
Tests fail intermittently in parallel mode. Suspect t.Setenv isn't compatible with t.Parallel. Please refactor to isolate env vars safely.
SUMMARY
Tried three approaches and confirmed t.Setenv can't coexist with t.Parallel. Resolved by moving env vars into table-driven test cases.
commit 9f3a8b1 · 2 files changed · +42 −18 · opened in 0.2s
INSIGHTS · REPORT GENERATION
Capture is just the start — sessions become reports.
One slash command and AIFlare turns your accumulated sessions into readable reports, saved to the console. From a single session to your team's week — five report types, organized by cadence.
Per session
- Session SummaryWhat was done and why, on one page
- Session CompareTwo sessions side by side — pivots, overlapping files
- Prompt EvaluationCoaching on prompt quality · before→after rewrites
Per day
- Daily DigestA day's commits, sessions, and tag mix (DEV / PM)
Weekly · team
- Weekly Team DigestYour team's week and key decisions · shareable via token (DEV / PM)
Weekly Team Digest2026-W17 · Team only
87
commits
8
contributors
3
key decisions
Notable decisions this week
→ Isolate env vars in parallel tests — t.Setenv rejected→ Switch webhook retries to exponential backoff→ Deprecate legacy /v1/charges endpoint
Distribute to the team via share token →
WHY AIFLARE
You don't start every session from zero.
AIFlare records every commit together with the AI agent session that produced it as a single entry — code change and intent aren't separated. Not a side archive, but context attached to the code.
02
"Where does this context live?"
Stored safely in AIFlare's backend. Org · Team · Project permissions control exactly who sees what, and the console gives you a single screen to search across them.
03
"Why do we repeat the same mistakes?"
Because agents start every session from zero. Context Inject replays the intent, decisions, and rejected alternatives from an earlier session into the next one with a single sessionId line.
RECALL · BACK TO THE AGENT
Captured context, back where you write code.
Context in AIFlare doesn't just sit in a dashboard. Slash commands and MCP tools feed intent and decisions straight back into the agent session — so you continue from the record, not from guesswork.
/why
Why is this line written this way?
Ask why a specific line exists while debugging. AIFlare runs git log -L to find the commits that produced it, and returns each commit's intent, rejected alternatives, and original conversation as a timeline. The "why" git blame can't give you.
why
$ /why src/auth/TokenService.kt:42
BUGFIXBlock refresh-token replay attacks
INTENT
If the same refresh token is used twice, treat it as theft and revoke the entire token family.
Diff around this line
+ if (token.used) revokeFamily(token)/context-inject
Last session, into this one.
Inject a previous session's saved summary report into a new agent session's context with a single sessionId line. Pick up work you left days ago — or a teammate's decision trail — without re-explaining it from scratch.
context-inject
$ /context-inject 2026-05-14-a3f9b201
## Injected Session Report — payment retry logic
Intent · 2 rejected alternatives · 3 changed files
✓ Injected into the current session — continuing from here.
Start recording intent today.
AIFlare is free during open beta. No card, no limits on team size. Install once and you'll never lose the context behind a commit again.
skill
curl -fsSL https://aiflare.dev/install.js -o install.cjs && node install.cjs && rm install.cjs