---
source_hash: ec46b0dce61f1f3810a8f5229f69c1d1284512ea449abae0460008d560b44134
post: using-it
folder: ari-os
shape: doctrinal
---

# Using ARI-OS day to day — Context Pack (for agents)

Agent-facing context pack for the published post `using-it`. Needles below are
copied **verbatim** from the `.mdx` body (this page's literals live in the body
itself, not in a composing component). Reproduce them exactly — do not
paraphrase, re-case, re-hyphenate, or "fix" underscores vs hyphens. When the
post and this pack disagree, the post wins: re-fetch
`content/posts/ari-os/using-it.mdx` and re-extract.

The load-bearing convention throughout: the **product is `ARI-OS`** (hyphen,
caps), the **Python module is `ari_os`** (underscore). Every CLI below uses the
underscored module path. Do not normalize either direction.

---

## 1. One-line gist

ARI-OS's daily surface reference: the **slash commands**, **skills**, **Cortex
memory MCP tools**, **dispatch**, the **monitor**, and the **control panel** —
the exact verbs, tool signatures, flags, and subcommands you drive every day
from one chair while workers run in the background.

---

## 2. NEEDLES (verbatim — do not confabulate)

### Frontmatter (exact)
- `slug: using-it`
- `title: "Using ARI-OS day to day"`
- `folder: ari-os`
- `type: how-to`
- `status: published`
- `date: 2026-06-15`
- `summary: "The commands, skills, Cortex MCP tools, and the dispatch and monitor surface you drive every day."`
- `related: [overview, install, brain]`

### Slash commands (10 — run inside the main session; name is the exact token)
- `/brainstorm` — "Open a structured brainstorm that pulls intent and requirements out before any code is written."
- `/dispatch` — "Hand a brief to a background worker and keep working in your own session."
- `/dream` — "Run a memory consolidation pass, distilling raw capture into durable signal."
- `/handoff` — "Write a clean handoff document so the next session can pick up where you left off."
- `/monitor` — "Open the live monitor to watch every running worker in one place."
- `/morning` — "Run the morning briefing: a maintenance check and a dashboard of what is in flight."
- `/night` — "Close the day out with an end-of-day review layered on top of a consolidation pass."
- `/prefetch` — "Warm the Cortex ahead of a task so the right context is loaded before you ask."
- `/recall` — "Pull the most relevant past context for a query straight into the conversation."
- `/remember` — "Save a fact, a decision, or a piece of thinking to the memory layer."

### Skills (10 — orchestrator can step into these by name)
- `advisor` — "Advisor-driven development: the orchestrator plans, cheaper models execute, and complexity escalates back up."
- `brainstorm` — "Explore intent, requirements, and design before any building begins."
- `dream` — "Consolidate memory: orient, gather, distil, and prune."
- `handoff` — "Produce a structured handoff so context survives the end of a session."
- `morning` — "Assemble the morning briefing from maintenance state and active work."
- `night` — "Run the end-of-day strategy review and consolidation."
- `prefetch` — "Load the context a task will need before the task starts."
- `recall` — "Retrieve relevant memory across the brain for a query."
- `remember` — "Commit text to the shared memory layer."
- `teach` — "Capture your methodology, patterns, and operating principles into memory drafts as you work."

### Cortex MCP server
- MCP server name: `ari-os-cortex` (hyphenated — the server, not the module)
- Manual registration (exact, note `python` not `python3`):
  `claude mcp add ari-os-cortex -- python -m ari_os.tools.cortex.mcp_server stdio`
- Module path: `ari_os.tools.cortex.mcp_server` (underscored) — wired up by the install guide.

### Cortex MCP tool signatures (10 — exact, including defaults in parens)
- `brain.recall(query, mode='default', k=12, token_budget=4000, session_id)` — "Recall the most relevant past context for a query."
- `brain.remember(text, layer='semantic', source)` — "Store a memory. It dedupes by the sha256 of the text."
- `brain.lineage(chunk_id)` — "Trace where a memory came from."
- `brain.regions()` — "List the memory regions."
- `brain.tracts(n=100)` — "List the strongest connections."
- `brain.entities(kind, k=50)` — "List known entities."
- `brain.relations(entity, k=50)` — "List an entity's relations."
- `brain.modes()` — "List the retrieval modes."
- `brain.see(image_path, mode='visual')` — "Describe an image. Off until LENS plus a local vision model are enabled."
- `brain.lens(slug)` — "Apply an aesthetic lens. Off until LENS is enabled."
- Default-off tools: `brain.see` and `brain.lens`. `brain.see` needs **LENS plus a local vision model**; `brain.lens` needs **LENS**.

### Dispatch — start a worker (exact command)
- `python3 -m ari_os.tools.dispatch start --executor sonnet --task-file BRIEF.md --cwd ./worktree --label mytask`
- Executors (only three, lowercase): `haiku`, `sonnet`, `opus` — "Pick the smallest one that can do the job, and escalate only when the work demands it."
- Access-shaping flags: `--add-dir` (grant an extra directory), `--read-only` (no write access).
- Hard rule: "Dispatch refuses a git repository root as `--cwd`. Point it at a worktree or a subdirectory instead, so a worker can never act on the whole repository at once."

### Dispatch — manage running workers (exact)
- `python3 -m ari_os.tools.dispatch list` — shows what is in flight.
- `python3 -m ari_os.tools.dispatch questions` — surfaces anything a worker is blocked on.
- `python3 -m ari_os.tools.dispatch answer <worker_id> --answer "..."` — sends your reply to a specific worker.

### Monitor
- Launch: `python3 -m ari_os.tools.monitor`
- URL: `http://localhost:7777` — "Leave it running while you work, and you can see progress, questions, and completions without switching context."

### Control panel — base command + subcommands
- Base: `python3 -m ari_os.tools.arios`
- Subcommands:
  - `keys` — manages the API keys for any models you use.
  - `theme <name>` — switches the visual theme.
  - `toggle <feature> on|off` — turns a feature on or off.
  - `update` — pulls the latest version.
  - `cortex <setting> [value]` — configures the memory brain; `setting` is one of:
    - `status`
    - `embeddings <auto|google|ollama|off>`
    - `mode <name>`
    - `wander on|off`
    - `ears on|off`
    - `lens on|off`
- Worked examples (exact):
  - `python3 -m ari_os.tools.arios cortex status` — check the state of the brain.
  - `python3 -m ari_os.tools.arios cortex mode focus` — switch the retrieval mode to focus.

### Diagram + pullquote
- Diagram `src`: `/posts/ari-os/dispatch-lifecycle.png`
- Diagram `alt`: `The dispatch lifecycle`
- Diagram `caption`: `A brief is dispatched to a worker, the worker runs in its own worktree, and it reports back to your session.`
- Pullquote (load-bearing thesis sentence): `You direct from the chair. The workers run in the shadows.`

### Cross-references inside the body (exact hrefs)
- Overview: `/documents/ari-os/overview`
- Install guide: `/documents/ari-os/install`
- Brain page (cited twice): `/documents/ari-os/brain`

---

## 3. GIST (the argument condensed)

This page is the **operator's manual** for the daily surface, and it is built
on one repeated shape: **you stay in the conversation and set direction; workers
run elsewhere and report back.** Every section is an instance of that split.

Three layers of verbs are enumerated. **Slash commands** (`/dispatch`,
`/recall`, `/dream`, …) are what you type in the main session. **Skills**
(`advisor`, `brainstorm`, `teach`, …) are the named capabilities the
orchestrator can step into — a command usually invokes the matching skill, but
the orchestrator can also reach for a skill unprompted. **Cortex MCP tools**
(`brain.recall`, `brain.remember`, …) are what the model calls directly once
the `ari-os-cortex` MCP server is registered.

Dispatch is the delegation spine: one prompt becomes several background
workers, each in **its own worktree**, gated by three executors (`haiku` /
`sonnet` / `opus` — smallest that can do the job, escalate only when demanded)
and two access flags (`--add-dir`, `--read-only`). The hard safety rule is that
dispatch **refuses a git repository root as `--cwd`** — you must point it at a
worktree or subdirectory, so no worker can act on the whole repo at once. Once
running, the loop is `dispatch list` → `dispatch questions` → `dispatch
answer`, with the monitor (`localhost:7777`) as the one live window onto all of
it.

The control panel (`python3 -m ari_os.tools.arios`) is the single configuration
surface — keys, theme, feature toggles, updates, and the `cortex` subcommand
whose settings (`embeddings`, `mode`, `wander`, `ears`, `lens`) are exactly the
switches that change what the memory brain pulls and how. Two tools
(`brain.see`, `brain.lens`) are default-off pending LENS (and, for `brain.see`,
a local vision model), so an agent that assumes vision is available is wrong by
default.

---

## 4. KEY LEARNINGS

1. **The daily surface is one shape, repeated.** Direct from the chair, workers
   run in the shadows. If you find yourself executing grind-work inline in the
   main session instead of dispatching it, you are not using the surface as
   designed — route it through `/dispatch`.
2. **Commands, skills, and MCP tools are three distinct layers.** Commands are
   what you type; skills are what the orchestrator steps into; tools are what
   the model calls. They often share a name (`brainstorm`, `recall`,
   `remember`) but are not interchangeable — do not conflate `/recall` the
   command with `brain.recall` the tool.
3. **Pick the smallest executor, escalate only on demand.** `haiku`, `sonnet`,
   `opus` — in that cost order. Reaching for `opus` by default defeats the
   point of the layer.
4. **Never dispatch against a repo root.** `--cwd` must be a worktree or
   subdirectory; dispatch hard-refuses a git repository root so a worker can
   never act on the whole repository at once. This is a safety invariant, not a
   preference.
5. **Two Cortex tools are off by default.** `brain.see` and `brain.lens` are
   gated behind LENS (and `brain.see` additionally needs a local vision
   model). Do not assume image/lens capability exists until those are enabled.
6. **The control panel is the single config surface.** Keys, theme, feature
   toggles, updates, and every cortex knob (`embeddings`, `mode`, `wander`,
   `ears`, `lens`) live under `python3 -m ari_os.tools.arios` — there is no
   second place to configure ARI-OS.
7. **`brain.remember` dedupes by sha256 of the text.** Re-storing identical
   text is a no-op; to update a memory, change the text. This shapes how you
   should phrase `remember` calls.

---

## 5. FILE INDEX (load-bearing — re-fetch to verify needles)

Unlike the sibling `tools-and-plugins/ari-os` page, **this page's needles live
in the `.mdx` body itself** (it is prose + JSX block components, not a single
composing component). The literals to re-verify are:

- `content/posts/ari-os/using-it.mdx` — the post
  under pack. **Primary source for every needle above**: the `SurfaceList`
  items (commands, skills, tools), the `<Code>` blocks (MCP registration,
  dispatch start, dispatch management, monitor, control panel + examples), the
  control-panel subcommand/cortex-setting bullet list, the `<Diagram>` props,
  and the `<Pullquote>`.
- `components/blocks` — the `Code`, `SurfaceList`,
  `WorkerFanOutFidget`, `Diagram`, `Pullquote` components imported by the post
  (referenced as `@/components/blocks`). They render props verbatim; they do
  not author the strings.
- `components/ari-os/content.ts` — the shared
  `ARI_OS` object. This page does **not** render from it, but it confirms the
  module-path convention (`ari_os`, underscore) and the install/uninstall
  commands the install guide wires up. Cross-check the underscore/hyphen split
  here if a module path needle looks "wrong".
- Sibling posts in the same folder (cross-linked, not needle sources for this
  pack): `overview.mdx`, `install.mdx`, `brain.mdx`, `advanced.mdx` — the
  install page wires the `ari-os-cortex` MCP server; the brain page covers
  modes, embeddings, regions, and consolidation in depth.

External / asset refs to re-fetch if stale:
- `/posts/ari-os/dispatch-lifecycle.png` — diagram asset (verify it exists in
  the public posts tree at the path shown).
- `http://localhost:7777` — the monitor's bound port (verify it has not moved).

> Backfilled 2026-07-18 from the published body; not needle-extracted at thought-time.
