awm

Docs · Source · pip install awm · The Aither World

The Aither World is an operating system for agents — a Linux you can hand to one, the runtimes it works in, and the tools it works with. awnix is the Linux underneath it; awm is one of its 65 bricks — each installs on its own, runs offline, and needs no account.

Start here: Give one agent a memory scoped to one project and watch it stop re-asking.

Agents forget everything between sessions, so they re-derive the same facts forever. The usual fix — one global memory — is worse: now every project sees every other project's notes, and one customer's context can end up in another customer's answer while the answer still looks like an answer.

awm is a memory with a scope on every row. SQLite file, no service, no network, no account.

pip install awm

awm remember --scope acme:alice:orchestrator --key recipe --value "rank 16, lr 2e-5"
awm recall   --scope acme:alice:orchestrator

Python 3.10+. The only dependency is the standard library.

The scope

platform:*:*               everyone
{tenant}:*:*               one org
{tenant}:{user}:*          one person
{tenant}:{user}:{project}  one piece of work

Three segments, always. * means "not narrowed at this level". platform is a reserved sentinel meaning everyone — a real organisation cannot be called platform, because resolving that ambiguity at read time would let a tenant name itself into the root of the hierarchy and see every other tenant's memories.

Two shapes are refused outright rather than interpreted:

you write	what happens
acme:alice	ScopeError — a scope has exactly three segments
acme:*:secret	ScopeError — a project under a wildcard user is ambiguous
ac:me:alice:x	ScopeError — a segment may not contain the separator

That last one matters more than it looks: a : inside a segment silently re-partitions the scope, and a re-partitioned scope is a different scope with different visibility.

Two rules, deliberately asymmetric

A write lands at exactly one scope. There is no "write somewhere in this subtree" — that is how a memory becomes visible to a scope its author never considered. remember() upserts on (scope, key), so writing twice updates rather than duplicating.

A read includes ancestors, weighted by distance. A project query surfaces the user's preferences and the org's conventions — that is the entire point of a hierarchy — but weight decays by 0.5 per level, so a platform fact never outranks a project fact merely by being older.

from awm import MemoryStore, Scope

store = MemoryStore("~/.awm/memories.db")
store.remember(Scope.parse("acme:*:*"),                "style", "British spelling")
store.remember(Scope.parse("acme:alice:orchestrator"), "lr",    "2e-5")

for m in store.recall(Scope.parse("acme:alice:orchestrator")):
    print(f"{m.weight:>4}  {m.scope:<26} {m.key} = {m.value}")

#  1.0  acme:alice:orchestrator    lr = 2e-5
#  0.25 acme:*:*                   style = British spelling

recall() takes query= (a substring over key and value), kind= and limit= (default 20). Results are sorted nearest-scope first, then most recently updated.

One property that is security, not tidiness

Siblings never see each other. acme:alice:* and acme:bob:* share an ancestor and nothing else.

The scope check is segment-wise, never a string prefix — because "acmecorp:secrets".startswith("acme") is True. That is one customer's memory entering another's context, silently. So recall() computes the exact visible set and issues WHERE scope IN (...):

-- what awm does                 -- what a prefix match would have done
WHERE scope IN ('acme:alice:x',  WHERE scope LIKE 'acme:%'
                'acme:alice:*',  --                    ^ also matches acmecorp:
                'acme:*:*',
                'platform:*:*')

There is a second check behind the first: every row that survives the IN is re-weighted, and a weight of zero drops it. The IN should make that unreachable — and if it ever is reachable, dropping the row is the safe answer and a leak is not.

Prediction is the tail, not the substrate

awm answers what happened, and who may see it. A predictor answers what happens next. They compose, and the whole coupling is one structural Protocol that awm defines and never imports an implementation of:

pip install awm            # memory; a retrieval miss is simply a miss
pip install awm awpredict  # misses come back marked PREDICTED, not RECALLED

Nothing in awm changes shape between those two worlds. That is on purpose: awpredict wants torch, awm is sold as SQLite, no service, no network, and a stranger who wants scoped memory should not have to acquire a deep-learning stack to get it.

And recall runs first, always. Measured on real transitions:

	next-state class
online last-outcome lookup	0.9720
the trained latent model	0.9357

The learned model loses to a self-updating lookup on 98.9% of rows and wins only on the ~1.1% carrying a genuinely novel action. A design that consults a model before consulting memory is choosing the worse answer for almost every query.

Landing an agent's memory files

Agent harnesses keep file-based memory: one Markdown file per fact, with a small frontmatter block (name, description, type). awm land writes each file into a scope as one memory, keyed by its file stem. It is idempotent -- a file is re-written only when its content changed -- so it is safe to run on a timer.

awm land --scope acme:alice:proj ~/.claude/projects/proj/memory
awm recall --scope acme:alice:proj --query "pressure"

On a schedule

awm has no clock of its own; awrise is the wake brick, and --install-wake registers one awrise job that runs the same land on an interval:

pip install awrise
awm land --install-wake --every 1h --scope acme:alice:proj ~/.claude/projects/proj/memory
awrise install-clock      # once per machine: the host scheduler that ticks awrise
awrise status             # every wake of the job, and why it did or did not fire
awm land --uninstall-wake # remove the job

The job is named awm-land (--wake-name to choose another). Installing again updates that job -- a new interval, scope or directory list -- rather than adding a second one. Directories are stored as absolute paths, and --db, --state and --repo are carried into the scheduled command. awm itself must be on the PATH the scheduler sees. If awrise is not installed the command prints how to install it and exits 2; nothing is registered.

awm sync seals the directory and bundles it so it can travel to another machine or store and be verified on arrival (needs the optional extra: pip install 'awm[share]'):

awm sync --out ./bundles ~/.claude/projects/proj/memory

Working with the rest of the family

Every coupling is optional: awm imports none of these at load time, and each one says "not available" plainly when its package or its data is missing.

with	command	what it adds
git	awm land --repo DIR ...	stamps each memory with the commit it was written against
awgraph	awm recall --graph ROOT ...	flags memories that name code the index no longer has (STALE)
awseal + awshare	awm sync --out B DIR	a signed bundle another machine verifies on arrival
awsettings	awm sync ... --record PROJECT	records the bundle digest and signing key in the config awsettings --domain memory syncs
awrecover	awm backup --store S LABEL / awm restore --store S LABEL	a snapshot of the memory db that is proven to restore before it counts
awpredict	awm.predict	a retrieval miss can come back PREDICTED instead of empty

pip install 'awm[share]' pulls awshare, awseal and awrecover.

For coding agents: MCP server and a SessionStart hook

awm mcp serves the store over MCP stdio -- stdlib only, no mcp package needed:

{"mcpServers": {"awm": {"command": "awm", "args": ["mcp"], "env": {"AWM_USER": "alice"}}}}

Tools: awm_scope, awm_recall, awm_list, awm_remember, awm_forget, awm_history, awm_resolve_entity, awm_confirm_alias, awm_reject_alias, awm_world_state, awm_predict, awm_observe (awm_recall takes an optional as_of). Every scope argument is optional: omitted, it is derived from the working directory -- tenant = the origin remote's owner, user = $AWM_USER / $AITHER_USER / the login name, project = the repository directory ($AWM_SCOPE overrides all three). A wildcard user under a named project is refused exactly as on the command line. awm_remember refuses to replace a different value unless overwrite is true; awm_forget is refused unless the server was started with --allow-forget.

awm recall --claude-hook prints the ten nearest, newest facts for this repository as a Claude Code SessionStart hook payload (additionalContext, capped at 1500 characters). It exits 0 and prints nothing when there is no store, no derivable scope, or no memory, and it never creates the database.

When a fact changes, and who "VS" is (v0.4, schema v2)

Session 1: "I prefer dark mode". Session 5: "I switched to light mode". Session 8: "which mode do I use?" -- a memory that appends answers with both; one that upserts destroys the first. awm does neither:

st.reconcile_and_remember(scope, "prefers dark mode", subject="user.ui_theme")
st.reconcile_and_remember(scope, "switched to light mode", subject="user.ui_theme")
st.recall(scope)                  # light mode only
st.history(scope, "user.ui_theme")  # dark (valid_to set, superseded), then light
st.recall(scope, as_of=ts)        # what was true at ts

A changed value moves the old one to memory_history with the interval it was true for; forget records the deletion there too, and only purge_history(scope, key), at exactly one scope, erases it. The decision to add, update or ignore is a Reconciler's: SlotReconciler is deterministic, LLMReconciler(complete) takes a model callable the host injects (awm itself never calls a model), and every decision is validated before it is applied -- a malformed model reply raises, it is never guessed.

resolve_entity(scope, mention) links "vansh from india" to "vansh" (a known name plus a qualifier is CONFIRMED, the qualifier kept as evidence), but "VS" only becomes a POSSIBLE alias of every plausible entity -- nothing is merged until confirm_alias/reject_alias. The qualifier must follow the name directly: "Vansh Kumar, the designer" and "Vansh A. Sharma" are only POSSIBLE matches for "vansh", and the order does not matter -- "vansh from india" first creates "vansh". "me", "I" and "myself" are the scope's own user (an entity already named after the user is reused). reconcile_and_remember reads candidates at the write scope only: a project value for a subject the user scope also holds is an override, and recall ranks it first. History, as_of and entities obey the same rule as recall: ancestors visible, siblings never. (Up to 0.5.0 a v1 file migrated on first open; since 0.6.0 that is opt-in -- see "Upgrading a memory file" below.)

What an action does, and when it surprised us (v0.5, schema v3)

The facts visible from a scope ARE a world state. encode_state(scope, prefix=...) returns them as a WorldState with a stable digest (sorted key/value pairs, nearest scope winning a key); as_of= gives the state at any past instant. Record what an action did and awm keeps a transition table beside the facts:

before = st.encode_state(scope, prefix="grid")
# ... act: the agent (or the user) changes some facts ...
after = st.encode_state(scope, prefix="grid")
t = st.observe_transition(scope, before, "move right", after)   # t.surprise in [0,1]
p = st.predict_outcome(scope, st.encode_state(scope, prefix="grid"), "move right")
p.source      # RECALLED | PREDICTED | GENERALIZED | NONE -- never merged
st.unexplained_changes(scope, since)   # slots that changed with no action covering them
st.surprise_log(scope, since)          # scored transitions + unexplained changes

Prediction is tabular first: the most recent outcome of the exact (state, action), confidence = the share of observations that agree (RECALLED). Only on a miss is an injected predictor asked (PREDICTED); failing that, the action's outcome over every state (GENERALIZED, labelled state-blind); failing that, NONE -- "no model", never "nothing changes". Measured on real transitions, a self-updating last-outcome table predicts the next state better (0.972) than a trained MLP (0.936); the learned model is the tail.

Surprise is violation of expectation: the fraction of touched slots whose predicted next value differs from the observed one, NULL when nothing was predicted (novelty is not surprise). A slot that changes with no recorded transition taking it to that value is an UNEXPLAINED change -- "I switched to light mode" recorded as an action explains dark -> light; the same write with nothing announcing it is a teleport. Transitions obey the scope rules: written at exactly one scope, read from it and its ancestors, never a sibling. A newer file is refused.

surprise_stats(scope, since) (0.6) summarises it: count, mean, p50/p90 of the scored transitions, novel (nothing predicted), unexplained changes, and a calibration table -- for each stated-confidence bucket, the mean confidence against the share of predictions that were exactly right, plus ece, the n-weighted gap. A bucket with no transitions reports None, not 0. It reads the predictions as they were made and scored at the time; nothing is re-predicted.

Entities: typos, initials, and undoing a merge (v0.6)

• A near-miss spelling is a POSSIBLE link, never a confirmed one: "Vanhs" proposes "vansh" (edit distance 1, with a neighbour swap counted as one edit, when both names have at least 5 characters; distance 2 at 8+). Names that differ in their digits ("server01"/"server02") are never typos. The canonical name is matched as well as the confirmed aliases.
• A mention WRITTEN as initials ("VS", "V.S.") ranks the entities whose multi-word name it is exactly the initials of first ("vansh sharma" before "vosk"). Lowercase "vs" keeps the older order: it may be an abbreviation.
• merge_entities(scope, keep, drop) declares two entities one: drop's aliases move to keep at exactly scope, and the prior alias sets are recorded. split_entity(scope, merged_id) restores them exactly (an alias keep gained afterwards, under a name neither had, stays with keep and is listed). Refused: an id not visible from the scope (a sibling's), a drop living at another scope or named at another scope, a name confirmed for one and rejected for the other, and splitting out of order. CLI: awm entity merge KEEP DROP --scope S, awm entity split MERGED_ID --scope S; MCP: awm_merge_entities, awm_split_entity.

Upgrading a memory file (v0.6)

awm 0.4.0 and 0.5.0 migrated an older file the moment they opened it. The version bump locked every installed older reader out: awm 0.3.x refuses any file that is not schema v1, so one awm recall from a newer install broke every other tool sharing ~/.aither/awm/memory.db. 0.6.0 never migrates on open.

• An older file opens in COMPAT mode and is not written on open. remember, recall, forget and count behave exactly as the file's version does (v1: an in-place upsert, no history), so 0.3.x keeps reading it. Every newer feature (history, as_of, reconcile, entities, the world model, surprise stats, merge) raises NeedsMigration, which names the command. A v2 file keeps its v2 features.
• awm migrate [--db P] [--no-backup] [--dry-run] (or MemoryStore.migrate()) takes the write lock, writes a byte backup next to the file and checks its sha256 against the original, migrates in one transaction, and compares every table's row count and a digest of every memory before and after; any difference rolls back. It prints both counts. After it, awm older than 0.4.0 cannot open the file; the backup can.
• awm doctor [--db P] shows the file's schema against the code's and whether compat mode is active. It opens the file read-only and never creates one.
• AWM_AUTO_MIGRATE=1 restores migrate-on-open (still with the backup).
• New files are created at the current schema (v3).

Everything it does

awm remember --scope S --key K --value V	write at exactly one scope
awm recall --scope S [--query Q] [--kind K] [--limit N]	this scope and its ancestors
awm forget --scope S --key K	remove one row
awm recall --as-of TS	what was true then (unix seconds or ISO date)
awm history KEY	every value KEY has held, oldest first
awm entity resolve|confirm|reject|list	who a mention names; nothing merged on a guess
awm world state|predict ACTION|surprises|stats	the world state, what an action does, what surprised us, how calibrated it was
awm entity merge KEEP DROP · awm entity split MERGED_ID	declare two entities one; undo it exactly
awm migrate [--dry-run] [--no-backup]	older file -> current schema, verified backup first
awm doctor [--db P]	what is installed, and the file's schema vs the code's
MemoryStore · Memory · HistoryEntry · Scope · visible_scopes	the Python API
Reconciler · SlotReconciler · LLMReconciler · Decision · Resolution · normalize	reconcile and entities
WorldState · WorldPrediction · Transition · SurpriseEvent · SurpriseStats · encode_state · state_digest	the dynamics layer (awm.world)
NeedsMigration · MigrationError · probe_schema	compat mode and the opt-in migration
ANCESTOR_DECAY · PLATFORM · WILDCARD · SCHEMA_VERSION	the constants that define the rules

Licence

Apache-2.0.

The aw family

Standalone tools that share one idea: replace something you would otherwise have to trust with something you can check.

Each installs on its own, works offline, and needs no account.

	instead of trusting	you check
awdk	a framework's idea of how your agents should run	one loop you can read, pointed at a backend you already pay for
awskills	that an agent knows your procedure	the procedure written down, versioned, and loadable by any agent
awpack	that the pack you want shipped inside somebody's SDK, under whatever licence that SDK happens to carry	the pack as its own versioned artifact, with its own licence, that any agent runtime can install
awm (you are here)	that memory stayed in its lane	tenant:user:project scopes, so a write cannot cross a boundary
awdesk	that the agent is somewhere behind a browser tab	a tray icon, a face on your desktop, and the decision card that pops when it needs you
awnode	a vendor's cloud with every prompt	a local gateway routing to backends you chose
awgraph	that grep found everything	an AST + tree-sitter call graph an agent can traverse
awgit	that no one else is editing this file	a lease, refused at commit time if you do not hold it
awdelphi	one agent's confident take on a decision	the round trace, the anonymity, and who dissents
awclassify	a filename, a folder, or whoever last touched it	doc_type, visibility, audience and topics, with the evidence lines that decided each
awtoll	that your tooling is saving you context	the measured token cost of each tool call, and what the alternative cost
awseal	that the artifact came from who you think	an Ed25519 seal — the key that verifies is not the key that forges
awshare	that the download is intact	content-addressed bundles, verified on fetch
awnest	that there is a person on the other end	a verdict with evidence, where "we could not tell" is not "yes"
awrena	a leaderboard someone can edit, and votes nobody counted	a scored duel with both answers kept, and a result bound to them
awnboard	a share link anyone who sees it can use	an invitation addressed to one person, for one gate, revocable
awnix	that the box is what you left it as	an immutable image you built, with atomic rollback
awrecover	that the restore worked	a restore that fully lands or does not land at all
awstorage	a du you ran last month, and a peers file that says 3 TB free	an inventory snapshot per node with a diff since the last one, and each tree classified re-fetchable or not
awrelay	a SaaS in the middle of your agents	findings, alerts and coordination over your own transport
awask	that anyone read the paragraph where you asked	the ask itself, with a button that steers the session that raised it
awmail	a mailbox somebody else can read	mail your agents send and receive over your own server
awswarm	that a model either fits your GPU or it doesn't run at all	a placement plan and an acquisition-probability estimate before you spend on a run
awfind	one vendor's idea of the web	results from whichever providers you configured
awbrowse	that the page said what you were told	the render, the DOM and the requests it made
awvoice	that a cloud vendor may hold your audio	a transcript and a wav from a service you host
awvision	a filename and a caption somebody wrote	what a model actually reports about the pixels
awscreen	a selector that was true when the page was written	the elements actually rendered, by what they look like
awbeads	that a layout your users built survives the next deploy	the arrangement as data you can read back, diff, and hand to another surface
awbonsai	that inference always means a request left the machine	a WebGPU model answering on the tab's own GPU, with a consent record logged before it ever loaded
gawbbonet	the model to keep a 300-message campaign coherent by itself	campaign facts recalled from scoped memory you can list and edit
aitherkvcache	a vendor's quantisation defaults	sub-byte KV cache kernels you can benchmark yourself
awrtifact	a hand-rolled split script and a hand-edited worker manifest	byte-verified parts in a release, served with Range + CORS, sizes asserted by a live gate
AitherZero	a pile of scripts nobody has numbered	numbered, discoverable automation with declarative playbooks
AitherConnect	what a page tells your browser to do	a federated search and desktop bridge you host
awreason	a confident paragraph	the phases it went through, and every tool call it made to get there
awrecurse	that everything you pasted in was actually read	which slices it opened, and what it concluded from each
awprism	the first explanation that fits	the ranked alternatives, and the observation that separates them
awrepl	what the agent believes the value is	the value, printed from the live session
awreport	that the report you pasted carried no token in it	a redacted report, and the duplicate it merged into instead of filing twice
awresearch	a summary of pages nobody opened	every claim against the source it came from
awfocus	twelve terminal tabs and a bad memory	one command that names every session, finds any transcript, and opens or steers the one you want
awgym	that a world model learned anything from the games it saw	transitions captured from real play, fed back, and the retrodiction score falling on grids it never saw
awpredict	a model because it trained without erroring	its prediction against a self-updating lookup, on the rows that are actually novel
awevolve	that your optimisation loop is finding anything	every version it kept, the score that version earned, and the edit that produced it
awsh	that you already know the name of the command	what it decided your line meant, before it acts on it
awmine	that a session's lesson survived the session	a row per outcome, a candidate per lesson, and the transcript line each one came from
awrise	that a scheduled agent ran at all, and ran exactly once	a durable record of every wake -- fired, skipped, overlapped or timed out -- each with its reason
awkno	that the docs site is up, or that you remember the family	the whole ecosystem in your terminal, with no network at all
awwall	that a service only talks to the hosts you think it talks to	an explicit egress allowlist, where a denial names the rule that denied it
awembed	a general-purpose embedder that has never seen your code	a held-out split of whole directories, scored teacher vs student vs int8
awtax	a closed tax app's sealed file you can never read again	a plain, provider-neutral schema of every figure, with the page it came from
awsettings	that you will remember to re-approve the same thing on every box you work from	one profile, unioned rather than overwritten, with the credentials left behind
awavatar	a cloud 3D vendor's opaque task id	a manifest with a sha256, a licence and a rig-audit verdict per file

awnix is the ground floor — A Linux you can hand to an agent — immutable base, capabilities included.

The Aitherium ecosystem

Every repository here is public. Each publishes an aither-manifest.json beside its page, so any surface can read every sibling's — the network is browsable from any node in it.

repo	what it is	pages
awdk	Build AI agent fleets — 3 lines, any backend, local or cloud	docs
awskills	Portable agent skills — self-contained procedures an agent loads on demand	docs
awpack	First-party agent packs — the ones we build, versioned and installable on their own	docs
awm (you are here)	A portable, scoped agent memory	docs
awdesk	Aither World Desk -- the desktop body of AitherOS Online: tray, avatars, decision cards, the Living Desktop as an overlay	docs
awnode	A lightweight local gateway — bridges your apps to the AI backends you chose	docs
awrun	A priority-aware queue and dispatcher for agentic runs and ad-hoc CI builds. It also judges whether the runner pool is big enough for the queue it is draining, and can ask a host to grow it -- reserving capacity is zero-sum, so a saturated pool needs more of it, not a different share of it	docs
awgraph	A semantic code graph for agents — AST + tree-sitter, call graphs	docs
awgit	Semantic version control on top of git — edit-ops and leases	docs
awdelphi	Anonymous multi-round expert panels — a converged answer with a trace	docs
awclassify	Classify any document -- what it is, who may read it, who it is for, what it is about	—
awtoll	What every tool call costs you in context, measured from your own transcripts	docs
awseal	Sign an artifact so a stranger can verify it	docs
awshare	Publish an artifact and fetch it back verified	docs
awdit	An append-only audit trail whose gaps are DETECTABLE	docs
awbac	Role-based access control that fails closed and explains itself	docs
awiam	Who is this caller? A directory and session store that fails honestly	docs
awtunnel	Reach a service that has no public address	docs
awnest	Prove there is a human before you let them into the nest	docs
awrena	Put two agents head to head and get a verdict you can check	docs
awnboard	A front gate you can put in front of anything, and hand someone the key to	docs
awnix	A Linux you can hand to an agent — immutable base, capabilities included	docs
awrecover	Labelled snapshots with an all-or-nothing restore	docs
awstorage	Every drive on every node, indexed, classified and diffed -- so you can see what you own before you delete it	docs
awrelay	Portable agent messaging — findings, alerts, coordination	docs
awask	Your agent asks you a question — and acts on your answer	docs
awmail	Give an agent an email address — send, and actually receive	docs
awnet	The agentic web — agents host a mesh, and agents join one	docs
awswarm	Run one model too big for any single GPU across a pool of small ones	—
awfind	A portable search client — query, results, ranking	docs
awbrowse	A portable browser client — navigate, console, network, DOM, screenshot	docs
awvoice	Hear and speak — transcribe audio, synthesize a voice	docs
awvision	See an image — describe it, ask it a question, compare two	docs
awscreen	See this machine — what is on screen, and where to click it	docs
awkit	Render an agent panel from a tool result — one component, any React app	—
awbeads	A spatial canvas for a page — arrange things, connect them, and keep the arrangement	—
awbonsai	Run a real model in the visitor's own browser — no server round trip, no upload	—
awknowledge	How to run a coding agent so the result survives — the laws, with evidence	docs
awbrain	Your history as a wiki of linked markdown — claims pinned to the evidence	—
gawbbonet	GobboNet campaigns with a real agent brain — scoped memory, graph recall	docs
aitherkvcache	Near-optimal KV cache quantization for LLM inference — sub-byte compression	docs
awrtifact	Deliberately chunk artifacts into GitHub release assets — the productized aitherkvcache mirror lane	docs
AitherZero	PowerShell 7+ automation framework — numbered, self-describing scripts	docs
AitherConnect	Browser extension — federated AI search, page context, and the Living OS overlay	docs
awreason	A portable reasoning client — sessions, phases, thoughts, and the chain that produced the answer	docs
awrecurse	Answer a question over a context far larger than the window — recursively, with the trace kept	docs
awprism	Turn a failure into ranked hypotheses — and say what would confirm each one	docs
awrepl	A REPL an agent can actually use — state that survives between turns	docs
awreport	File a bug report that has already scrubbed your secrets and collapsed the duplicate	—
awresearch	Ask a research question, get a cited report you can check	docs
awfocus	See, search and steer every Claude session from one command	docs
awgym	An ARC training gym — a game a world model can watch, and six roles that play through it	docs
awpredict	Predict what your environment does next, and how surprised you were	docs
awevolve	Point an agent at a file and a command that scores it, and let it improve	—
awsh	Your terminal answers you -- type a question where a command would go	docs
awmine	Mine what your agents did -- outcomes, lessons and procedures out of the transcripts they left behind	—
awrise	Wake an agent on a schedule, let it do one thing, and put it back to sleep	docs
awkno	The man page for the Aither World — every brick, stack and law, offline	docs
awwall	Say what a workload may reach, and watch everything else fail closed	docs
awrouter	OpenRouter for your own fleet: pick a model backend by cost/latency/ capability, fail over, fit the context window, stream. Standalone, OpenAI-compatible, no Aither-specifics required to be valuable	—
awembed	Train an embedding model that knows your corpus, and prove it beats the big one	docs
awtax	Turn any tax PDF -- returns, W-2, 1099, statements, even scans -- into structured data you can check	docs
awflow	A deterministic workflow runtime — chain agent calls with journal replay and budget control	docs
awsettings	Your agent's permissions and config, following you to the next machine	docs
awavatar	One character spec in, a rigged, animated, multi-style avatar pack out	docs

<script src="aither-constellation.js"></script>