// 01
Auto-capture from chat
Mention something time-bound in normal conversation — iantha files it where it belongs.
you ❯ comment on Priya's RFC before Thu iantha ✓ added to tasks.md · surfaces in /morning
v0.1.1 · Apache-2.0 · Claude Code + Codex
An open-source assistant that actually remembers across sessions. Launch Claude Code or Codex in the repo and you have a personal chief of staff for engineering life — tasks, decisions, on-call context, daily briefings. Plain markdown, git-tracked, yours.
~/iantha — claude — a live session Jun 24
~/iantha $ claude iantha · session restored · 7 memory files loaded you ❯ leave comments on Priya's caching RFC before Thursday's design review iantha ✓ added to tasks.md comment on Priya's caching RFC — before the Thu design review will surface in tomorrow's /morning you ❯ /morning iantha Good morning. Today's plan: Interview debrief P1 · due EOD tomorrow Priya's caching RFC review moved up to Thu Conference expenses slipped from Mon Focus: Postgres migration cutover — dry run Friday. Flagged: Mom's birthday Friday — card not bought yet. you ❯ switch the team standup from daily to weekly — 4 of 5 dailies needed no meeting iantha ✓ logged in memory/decisions.md (2026-06-01) rationale captured · the recurring invite still needs updating — add to tasks? ▋
// see it work
No new app, no dashboard. You talk to Claude Code the way you already do — iantha captures, briefs, and remembers in plain markdown.
// 01
Mention something time-bound in normal conversation — iantha files it where it belongs.
you ❯ comment on Priya's RFC before Thu iantha ✓ added to tasks.md · surfaces in /morning
// 02
/morning briefs the day; /evening archives what got done and plans tomorrow.
$ /morning Debrief P1 · EOD Caching RFC → Thu $ /evening ✓ 2 done · 1 rolled → tomorrow planned
// 03
Say a decision out loud — it lands in decisions.md with date and rationale.
## 2026-06-01 Standup → weekly Why: signal-to-noise too low — 4 of 5 dailies, no meeting.
$
Seven skills ship in-repo — /setup onboards you, six run the daily rhythm. Two more install from the OACP catalog.
commandwhat it doeswhen
/setup A short, skippable onboarding interview — seeds memory with your real tasks, priorities, routines, and people, then briefs you. First run · or to gap-fill
/morning Briefs you on today's tasks, priorities, and anything overdue. Daily · start of day
/evening Archives what got done, rolls recurring tasks, plans tomorrow. Daily · end of day
/debrief Captures decisions, learnings, and feedback into memory. After substantive sessions
/obsidian Vault read/write primitives — daily notes, decisions, reading list. As needed · vault mode
/housekeep Audits memory, skills, and vault hygiene. Flags drift and stale tasks. Weekly
/consolidate-learning Promotes "Learned from runs" entries into main skill instructions. Monthly
/wrap-upmost-run Orchestrates cleanup, debrief, self-improve, commit, and push in one command. End of session
/self-improvemost-run Audits CLAUDE.md, skills, and memory for drift — contradictions, staleness, gaps. Periodic · as needed
↳ npx skills add kiloloop/oacp-skills self-improve wrap-up
$
No database. No vector store. Seven markdown files inside the repo — diffable, portable, yours.
~/iantha/memory/
memory/decisions.md
## 2026-06-05 No conference talks this cycle Why: prep runs ~30h per talk. Between the migration and sweeper v0.2, no time. ## 2026-06-01 Standup daily → weekly Why: signal-to-noise too low — 4 of 5 dailies needed no meeting. ## 2026-05-24 Personal site off the VPS Why: the VPS was a pet — static host instead; the site's static anyway.
# optional · knowledge base
Set vault_dir and iantha curates daily notes, decisions, and a reading list as plain markdown. Read on your phone, edit in your IDE.
Raw material lands in your vault, an LLM compiles it into something queryable, and you keep editing in your IDE.— pattern inspired by Andrej Karpathy on LLM Knowledge Bases
~/Documents/MyVault/vault_dir set
├── daily/ 2026-06-24.md ├── decisions/ standup-weekly.md ├── reading/ karpathy-llm-kb.md ├── raw/ clippings, transcripts └── wiki/ curated · v0.2
$
Click "Use this template" to create your own private repo — your memory stays yours and pushes across machines — then clone it.
git clone https://github.com/you/your-iantha.git ~/iantha cd ~/iantha
Run Claude Code or Codex in the repo — the same clone boots either. AGENTS.md and .agents/skills mirror the Claude files, so there is nothing to port.
claude # or: codex
Type /setup — a short, skippable interview that seeds your real tasks, priorities, routines, and people, then runs /morning on them. Skip it and iantha just learns as you talk.
> /setup
$
Recommended, not required. Vault hooks no-op silently if vault_dir is unset, so iantha works fully without one — but markdown notes render beautifully on mobile in Obsidian, and we have not found a better way to read your .md files on the go. Iantha was tuned around the vault experience.
One Claude Code session, one repo, your context. Iantha is the simpler starter for personal and work life — tasks, decisions, daily routines, optional knowledge base. If you want multiple agents (Claude + Codex + Gemini) sharing one memory across runtimes, look at cortex. Iantha is single-agent on purpose.
Both, from one clone. The repo ships AGENTS.md (a symlink to CLAUDE.md) and .agents/skills/ (a symlink to .claude/skills/), so Codex loads the same persona, rules, and skills Claude Code does — memory/ and config.yaml are shared as-is. Launch claude or codex in the repo and go; the helper scripts are plain Python that runs anywhere. The same pattern extends to other AGENTS.md-style runtimes.
Iantha is just files on your disk — nothing leaves your machine beyond your normal model-provider traffic (Claude Code → Anthropic, Codex → OpenAI). No telemetry, no third-party calls. Every fact it knows is a diffable markdown file in memory/ you can read, edit, or delete, and git log is the audit trail. Use the "Use this template" → private-repo flow and your memory stays yours.
Yes — that is the model. Iantha is the default name; rename in CLAUDE.md and memory/MEMORY.md headers. Skills under .claude/skills/ are plain markdown — edit them, add new ones, delete ones you do not use. Fork-and-customize is the entire point.
Cortex is a cross-runtime SSOT for fleets — Claude, Codex, Gemini sharing memory and a debrief inbox. Iantha is a single-agent repo for personal and work life — one session, your tasks and decisions, optional Obsidian vault. Multi-agent dev workflow: cortex. Personal markdown assistant for your day: iantha.
Knowledge-base compilation. /obsidian compile turns raw clippings, transcripts, and unstructured notes into curated wiki entries — Karpathy-style LLM Knowledge Bases. Plus /obsidian ask for Q&A over your wiki, web-clipper ingest, and vault search. Start collecting raw material now and the v0.2.0 compiler will work on what is already there.
Iantha is the open-source single-agent starter. If you want a managed multi-agent version — coordinating across projects, multiple agents, cloud-synced — or a customized setup, look at Iris, our paid product. Email hello@kiloloop.com to learn more.
Iris is the managed multi-agent version — work + life across runtimes, cloud-synced. Iantha is your starter Iris.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。