awmine

Source · pip install git+https://github.com/Aitherium/awmine.git · 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; awmine is one of its 65 bricks — each installs on its own, runs offline, and needs no account.

Start here: Point it at one transcript directory and read back what those sessions learned, with the line each lesson came from.

Turn the agent transcripts already on your disk into redacted rows of what failed, what you corrected, what you repeat, and what it cost.

Every coding-agent session leaves a JSONL transcript behind, and those files already hold the answers to four questions nobody re-reads them for: which tool calls fail and in what shape, where the agent claimed "done" and you said otherwise, which step sequences you run again and again, and what each session cost in tokens. awmine reads the transcripts as a stream, carries the state that spans lines across runs, and emits one row file per question. Every string in every row is redacted before it reaches disk; a row that fails redaction is a failed run, not a warning.

Standard library only. No daemon. The wake is whatever scheduler you already have.

Install

pip install awmine

Python 3.10 or newer. Linux, macOS and Windows. If awtoll is installed its transcript reader and shape rules are reused; if not, a copy of the same idiom is used so the two never disagree about what a step is called.

Quick start

awmine run --since 7d              # mine new bytes since the last run (default roots: ~/.claude/projects)
awmine report                      # counts, cost per project, top lessons and procedures, redaction hits
awmine export --harvest            # exports/harvest.jsonl -- training-example rows
awmine export --codex --top 10     # exports/codex_candidates.yaml -- lesson candidates (reads a backlog, never writes it)
awmine export --teach              # exports/teach.jsonl -- {fork,state,answer,reward} rows
awmine export --skills             # exports/skills/<name>/SKILL.md drafts + toolpack_candidates.json
awmine share                       # OPT-IN: nothing leaves the box without --share or AWMINE_SHARE=1
awmine share --share               # render qualifying procedures as awskills/candidates/<slug>/SKILL.md
awmine --self-test                 # proves every extractor finds its planted case and that redaction can FAIL

Output root is $AWMINE_OUT, default ~/.aither/awmine/ (created 0700). Roots come from --roots, $AWMINE_ROOTS, $AWTOLL_TRANSCRIPTS or the default, in that order.

What it emits

file	one row per	what the row says
outcomes.jsonl	paired tool call	`{fork, state:{tool,args_shape,cwd_kind}, answer:"yes"
lessons.jsonl	correction / interrupt / retry that worked	a claim, the human line that contradicted it, a marker score, quoted lines capped at 320 chars
turns.jsonl	ordinary next-task prompt after a claim	same row shape, kind turn -- kept apart so "lessons by kind" counts only real corrections
procedures.jsonl	maximal step window seen in >= 2 top-level sessions	normalized steps, n_sessions, n_occurrences, n_subagent_occurrences, success_rate, examples
cost.jsonl	session or subagent	tokens summed ONCE per API message id, tool-outcome ladder, API errors by status, last cost-state, awtoll's per-record delta

Every row carries ts and source: {root, path, line, session_id, top_session_id, ...} where root is <HOME>-substituted and path is relative to it. Three forked subagents of one run map to their parent's top_session_id, so they count as ONE session for procedures and are reported separately as subagent occurrences.

Shapes, not arguments

A Bash step is the binary plus at most one subcommand token ($ git push, $ pytest, $ python -c (inline)); any other tool is its name plus its input KEY names (Edit(file_path+new_string+old_string)). Flags, paths, hosts, URLs, e-mails, branch names and every other bare token are dropped, not placeholdered -- there is nothing to redact because nothing was copied. The operator denylist is applied to every step afterwards as belt and braces.

Redaction

One writer function redacts every string in every row (recursively) before bytes hit disk. The vocabulary: generic key/token/secret patterns, GitHub/OpenAI/Anthropic/Google/HF/AWS key shapes, JWTs, Bearer values, X-API-Key/X-Internal-Key headers, credential assignments, private-key blocks, bare hex of 32+ characters, home directories (a Windows user-profile prefix, /home/..., /Users/... become <HOME>/), e-mail addresses, and an operator denylist of terms read from $AWMINE_OUT/denylist.txt, $AWMINE_DENY and --deny (case-insensitive, word-bounded, each hit becomes [REDACTED:name]). The report prints the denylist term COUNT so an empty denylist is visible.

A credential assignment is matched with the keyword as the suffix of a longer identifier, because that is the shape a transcript is full of: DB_PASSWORD=, PGPASSWORD=, GITHUB_TOKEN=, AWS_SECRET_ACCESS_KEY=, export MY_API_KEY=. The space-separated flag form (--password <value>), -u user:pass and scheme://user:pass@host are covered too. The right-hand boundary is the =/: itself, so secretary=jane and http://host:8182/path are left alone.

Identifiers that are hex by construction (content_hash, input_sha, error_key, id, tool_use_id, session_id, promptId, ...) are exempt from the secret patterns; the exemption list is a module constant printed by --self-test, and the residual scan skips exactly those fields and nothing else. An exception inside redaction aborts the run with exit 2 before that file's rows or manifest entry are written.

The leak check has two halves, and says which is which

• Residual scan -- re-apply the vocabulary to every file awmine wrote. Any hit is exit 1. This is a self-consistency check: it shares the vocabulary's blind spot by construction, so it can prove the writer redacted, never that the vocabulary was complete. The run summary says so on the line that prints it.
• Vocabulary-free suspect scan -- an entropy/shape heuristic that shares no pattern with the vocabulary, so a credential-shaped assignment under a keyword the vocabulary has never heard of is still named (by its NAME, never its value). It excludes the shapes measured as false positives on a real corpus -- a markdown code span, a UUID, a filename, a lowercase slug, an id range, an ISO date, a hex hash -- and those exclusions are module constants, not a list that grows per finding. A warning by default, because a heuristic over a whole corpus that floods gets switched off; AWMINE_SUSPECTS_FATAL=1 makes it exit 1.

Both run over every file under the output directory -- rows, procedures, exports of any format, manifest.json and the steps/ cache -- not a hand-kept list of some of them. The manifest carries quoted transcript text (the last claim, open tool errors) and is keyed by the transcript's project directory, so it is redacted like anything else. The one file skipped is the operator's own denylist.txt, which is input. A file awmine wrote that cannot be parsed is exit 2, never a pass.

Changing the denylist after rows exist changes how paths are written in new rows; re-mine with --full from an empty output dir when you add a term.

Idempotency

manifest.json records, per (root index, relative path), the byte offset and line count reached plus a small ResumeState (pending call shapes, the last claim, open errors keyed by input hash, an open interrupt) holding only shapes, hashes, line numbers and redacted capped quotes. A re-run seeks to the offset and appends only rows from new lines; a call mined on one run whose result lands on the next pairs normally. A file that shrank or was replaced re-mines from 0 after its old rows are dropped from every row file (tmp + fsync + replace). Rows that a crash left past the manifest's line count are dropped and re-appended once. The manifest is replaced atomically only after a file's rows are on disk. procedures.jsonl and the report are recomputed from the steps cache on every run.

Rows leave the miner mid-file, not at EOF: the buffer is flushed every AWMINE_FLUSH_ROWS rows (default 4,000) so what is held does not track the size of the largest single transcript. Rows may lead the manifest and never lag it, which is the same invariant crash repair already relies on; if a transcript vanishes mid-read after a partial flush, its rows are rolled back to the last durable line, because the manifest will never name that file for crash repair to plan for.

The per-file counts block (conversation/sidecar/other/unreadable records and what was skipped) accumulates across incremental mines, like its siblings rows and unreadable_lines. A re-mine from 0 restarts the tally.

Exports

• --harvest writes rows with the field names of a training-harvest example (messages capped at 2,000 characters, content_hash = md5 of the WRITTEN messages, contains_private_prompts true iff redaction hit that row).
• --codex writes lesson candidates in a {id, source, title, origin, evidence, status, seen} row shape whose evidence always carries measured, an ISO date and a line count. It reads a backlog file READ-ONLY (--backlog, or $AWMINE_CODEX_BACKLOG, or a project default found under the nearest repo root) to skip ids already present, emits at most --top N (default 10) and reports emitted / already_present / held_back. It never writes the backlog.
• --teach writes {fork, state, answer, reward} rows. --post sends each as {domain: "decide.awmine.tool_outcome", state: <json string>, answer, reward} to $AWRISE_DECIDE_URL/decide/outcome with $AWRISE_DECIDE_TOKEN as a Bearer, 3 s timeout, any failure counted and never raised. Default is file only.
• --skills writes exports/skills/<name>/SKILL.md drafts (frontmatter name + description, numbered steps, measured counts; --awskills-form for the flat allowed-tools form) and exports/toolpack_candidates.json listing awmine_proc_<slug> functions. Drafts are capped at --skills-top (default 25, most sessions first) and the number held back is reported -- a real corpus yields tens of thousands of qualifying windows. Nothing is written outside $AWMINE_OUT.

Share what qualified -- opt-in, gated, one PR

A mined procedure is not done when it is a row. It is done when it is a pack somebody else can install. awmine share is that lane, and it is the one command that writes OUTSIDE $AWMINE_OUT, so it is the one command that is off unless you ask for it.

awmine share                         # OFF: writes nothing, says so, exits 0
awmine share --share                 # this run only
AWMINE_SHARE=1 awmine share          # an unattended wake
awmine share --share --repo <checkout> --top 10

What a run does, in order:

1. Selects the procedures the miner's own signal gate already qualified (>= --min-sessions top-level sessions, default 2, most-repeated first, --top per run, default 10). A window of one step repeated is skipped -- a row file outlives the code that wrote it.
2. Renders each one to awskills/candidates/<slug>/SKILL.md: frontmatter name + description, the numbered steps, the measured counts, and the source path:line span of every example. The slug is a pure function of the procedure, so a re-run updates a candidate instead of creating a second one.
3. Redacts again. The rows were redacted when they were mined; the rendered text is redacted a second time here, and a candidate whose text still CHANGES under redaction is refused outright rather than quietly scrubbed -- a second hit means something got past the first pass.
4. Asks the publish gate. Every candidate is staged into a throwaway pack and put through check_skills_publishable --offline as a subprocess (--gate, $AWMINE_PUBLISH_GATE, or the checker found in the checkout). A candidate it rejects is not written. A finding that belongs to no candidate stops the whole run, and a gate that cannot answer is exit 2.
5. Prints what you do next -- add, commit, open ONE pull request -- and does none of it. awmine never commits, never pushes, never opens a PR and never touches the network here. The repository's existing skills-sync workflow is the publisher; a second lane is how mirrors drift.

--repo (or $AWMINE_SHARE_REPO) names the checkout that carries the pack; a named tree without one is an error, never a reason to write somewhere you did not name. Exit 0 wrote (or opted out), 1 every candidate was refused, 2 there was nothing to judge with.

Wake it

awrise add --name awmine-mine --every 6h --run "python -m awmine run --since 7d" --timeout 1800
awrise add --name awmine-share --every 24h \n  --run "python -m awmine run --since 7d && AWMINE_SHARE=1 python -m awmine share" --timeout 1800

Register it on the host that holds the transcripts. The wake never passes --post; teaching a decision door stays a deliberate awmine export --teach --post.

Exit codes

0 clean · 1 a measured failure (a row failed redaction or validation, a residual hit, a leak suspect under AWMINE_SUSPECTS_FATAL=1, a self-test check failed) · 2 could not judge (no transcripts found, manifest unreadable, a written file unparsable, output dir unwritable, redaction raised). Never 0 on silence.

Environment: AWMINE_OUT output dir · AWMINE_DENY extra denylist terms · AWMINE_FLUSH_ROWS rows buffered before a mid-file flush · AWMINE_SUSPECTS_FATAL=1 make the vocabulary-free suspect scan a gate · AWMINE_SHARE=1 opt in to the share lane · AWMINE_SHARE_REPO the checkout to share into · AWMINE_PUBLISH_GATE the gate to run before writing.

Self-test verifies

• run_over_fixture_exits_0_with_0_residual_hits
• outcomes_denied_call_is_yes_and_retry_is_no
• outcomes_interrupted_call_is_unpaired_not_a_row
• outcomes_state_holds_only_tool_args_shape_cwd_kind
• bash_shape_drops_hostname_url_branch_email
• lessons_correction_interrupt_retry_worked_each_found
• correction_scores_at_least_one_whole_marker
• ordinary_next_task_lands_in_turns_not_lessons
• retry_worked_records_distance_and_count
• evidence_carries_measured_iso_date_and_line
• procedure_window_emitted_once_and_maximal
• three_forked_subagents_count_as_one_top_level_session
• success_rate_reflects_failing_occurrence
• no_bare_argument_survives_in_any_step
• cost_one_row_per_session_and_subagent_journals_excluded
• usage_counted_once_per_message_id_not_per_record
• api_error_counted_and_last_cost_state_kept
• awtoll_per_record_sum_exceeds_per_message_sum
• second_run_emits_zero_rows_and_identical_manifest_bytes
• split_call_across_runs_pairs_as_ok
• split_claim_across_runs_yields_correction
• remine_from_zero_leaves_row_counts_unchanged
• crash_between_flush_and_manifest_repaired_without_duplicates
• harvest_content_hash_is_md5_of_written_messages
• contains_private_prompts_true_on_every_redacted_row
• teach_rows_are_exactly_fork_state_answer_reward
• prefixed_credential_assignment_is_redacted (DB_PASSWORD=, --password <v>, -u user:pass)
• relaxing_that_anchor_does_not_redact_ordinary_text
• leak_sweep_opens_manifest_and_steps_cache_and_skips_the_denylist
• suspect_scan_names_a_shape_the_vocabulary_does_not_know
• suspect_scan_is_silent_on_awmines_own_output
• share_is_off_without_the_flag_and_writes_nothing
• share_renders_a_gated_candidate_with_its_marker_steps_and_source_spans
• share_slug_is_stable_across_runs
• share_refuses_a_secret_that_survived_into_a_row
• no_literal_canary_in_any_file_written -- manifest and steps cache included; exits 1 under AWMINE_SELFTEST_BREAK=redact and =denylist

Privacy limits, stated

Attachment payloads (hook stdout, prompt snapshots), raw tool inputs and whole transcripts are never copied anywhere, including internal state. Lesson quotes are capped at 320 characters and export message bodies at 2,000. Name detection beyond the e-mail pattern and the operator denylist is not attempted; an undetected customer name inside a quoted line is the residual the denylist exists to close.

License

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	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 (you are here)	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	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 (you are here)	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>