The LoRA is not the textbook. It is the specialist who knows how to use the library.

One GPU, one small resident model, and a shelf of QLoRA adapters — each an expert defined by its corpus. An OpenAI-compatible API reads every request, serves it with the expert whose corpus it falls in, and sends everything else to a frontier model. The client names no model.

Version 1.0 adds what a small expert cannot hold in its weights: a library of short notes per subdomain, and an adapter trained on the habit of navigating it. The weights hold the navigation; the notes hold the content, and a person can read, diff and correct them.

Read the architecture article → En español The memory Repository

A reader at a desk before a wall of card-catalogue drawers labelled Procedures and Encyclopedia; a small radar dish on the desk lights three drawers; through an open door, a domed building on a hill labelled frontier. — The specialist, the library, the radar that lights the drawers worth opening — and, through the door, the frontier for everything else.

The finding that reorganised the project

11 of 90 became 90 of 90

Same adapter, same cases. Only the serving path changed.

A fluid-mechanics expert — 6-to-9-step chains with a calculator and a per-case handbook — scored 11/90 when tool results came back as tool_calls messages. Written inline, the way its corpus taught it to read them, it scores 90/90 (79 : 0, paired). For four days that gap had been read as “small models cannot reason”.

An expert is its corpus

Tool block, argument order, system prompt, result format.

Under a foreign system prompt 2 of 32 live turns call a tool; under its own, 19 of 32. Serve an expert the way its corpus taught it, and suspect the path before the model.

The pool moves bases and holds

Same corpora, same recipe, a new base.

Both released members were retrained on Qwen3.5-4B and each ties its own earlier release: inbox triage 471/475, desk commitments 240/240. They have since moved again: every released member is now on Gemma 4 E4B, and the Qwen releases stay as the control arm.

What stays text

What deliberately never becomes a weight?

The library. A weight delta cannot be read, cited or corrected by a person; a half-page note can. Site rules override the textbook before the expert ever sees the note.

The core of version 1.0

The five parts of version 1.0: experts, router, memory, runtime and the release contract. — Five things: experts, a router, the memory, a runtime that referees, and a release contract that nothing ships without.

A walk through the memory: a question, a search that lights a few notes, notes opened one after another, a calculation, an answer. — Three verbs — `search`, `open`, `calc` — with results written inline. A software referee turns the pages, applies the site's rules and cuts a walk that skips a required step.

built — measured in part All five parts are built. Libraries now exist over an open textbook’s procedures (CC BY 4.0) and over US regulations ingested verbatim. On real documents the central claim holds where it was measured: an expert trained on walks over one regulation family cites on a family it never saw, and a number edited in the library after training is answered from the edit (below). On generated procedures it did not: the library arm did not beat the untrained base handed the same notes.

What is measured

what	the number	run
A reasoning expert served the way its corpus taught	90/90 against 11/90 through `tool_calls`, 79 : 0 paired	M7 arm 0b
Inbox triage, retrained on `Qwen3.5-4B`	471/475, a tie with its earlier release; the bare base scores 346/475	M1
A second expert on the same inbox, a different question	240/240, a tie with its earlier release	M1
The API routes per request; the client names no model	0.546 → 0.775 on a 240-case replay, 0 misrouted	P41, P62
A procedure merely pasted into a 3B’s prompt is not followed	0 tool calls on 351/351; the trained expert beats it 137 : 1	P61
Trained on walks over real regulations, an expert cites on a regulation family it never saw	18/23 against the untrained base’s 9/23, 10 : 1 paired; a second seed ties it	REAL3
A number edited in the library after training is answered from the edit, not the weights	17/17 rows answer the new value, each cited to the edited statement, 0 stale	EDIT0
The router that replaced the keyword dictionary as the default: a request is factored into its task and its content	foreign text served locally 0/600 against the dictionary’s 294/600; unseen senders 0 of 120 lost	ROUTE0
A long-policy, multi-turn agent benchmark (τ²-bench airline, 20 test tasks × 4 trials), untrained bases — a headroom check, no member trained yet	Gemma 4 12B pass^1 0.45 against the E4B’s 0.175, +27.5 pp, 10 : 0 tasks; the 31B’s 0.5375 is +8.8 pp over the 12B, inside the noise	T1, T1b

What is not, or does not work

No real traffic yet. Suites are generated, or built from real documents ingested verbatim; live runs go through a real agent runtime, on synthetic organisations.
The router is literal by design. A paraphrase of a member’s task is sent to the frontier, not served locally (0/120 kept local), and two learned routers — n-grams and embeddings — both lost every request from a sender they never saw.
The radar that picks which notes to open has not passed its gate, and the library arm lost its kill test on generated procedures.
Not installable yet. There is no package or container, and the serving code does not yet load the per-role packs it lints. What exists is a research codebase whose runs are reproducible on Colab.
The saving in money is not established. The frontier side of a replay has been costed; the local GPU has not, and none of it is real traffic.

The full ledger, including everything that failed and the instruments that lied: RECORD.md. The position: PLAN.md.

Where it sits in an organisation

One expert per role

Under the agent column, not beside it.

An organisation that runs on agents draws the same picture: people in a few roles, an agent runtime with one agent per role, the applications they operate, the channels people already use, and one database underneath. In that picture every agent is a system prompt over the same remote model. Here each role becomes an expert — an adapter trained on how this organisation does that job — and the role a message arrives from is the route.

Three places for three things

Records, habits, knowledge.

Records stay in the database. Habits go in the adapter. Knowledge stays in notes a person can read and correct — two drawers per role: how we do it here, and what we know. What an expert is measured to handle is answered on the organisation’s own machine; the rest goes to the frontier, or to a person where policy says nothing leaves the building.

A solution architecture in five layers: people in four roles; an agent runtime with one agent per role; order and back-office applications; app and messaging channels; one database with identity, payments and monitoring. Under the agents, one graphics card drawn as a bookshelf: one thick spine, the resident model, and a thin spine per role, each with two drawers of notes. A signpost routes by role; dashed lines leave for the frontier and for a person. — A distributor, as the example: one adapter per role under the agent column, two drawers of notes under each, the role as the route — and the systems of record left where they are.

organisation	roles that become experts	what goes in the two drawers
a distributor	customer service, receiving, dispatch, purchasing, claims	the site’s handling procedures · catalogue, carriers, service levels
an accounting or law office	intake, document review, deadlines, billing	the firm’s checklists and templates · the rules of its jurisdiction
a school or training centre	enrolment, teaching support, communications, purchasing	how this school handles each case · programme, calendar, regulations
a repair or field-service company	equipment intake, diagnosis, spare parts, warranties	the procedure for each kind of repair · manuals, parts lists, warranty terms
a club or community centre	memberships, activity sign-ups, facilities, collections	how this club handles each case · activities, fees, house rules
a property manager	tenant requests, maintenance, collections, suppliers	the escalation procedure per building · contracts, by-laws, supplier terms

not measured These are where the design points, not results. What they share is what makes a region worth an expert: the same few procedures, repeated daily, with local rules that differ from the textbook, over data that should not leave. The first library in the repository is built from an open textbook’s step-by-step procedures.

For a team running agents together

Four teams and their direct messages flow through an agent runtime and a proxy to one GPU holding one small resident model and an adapter per team, each with two drawers: how this team does it, what this team knows. A dashed line leads to the frontier as fallback. — Many teams, one machine. The group a message arrives from is the route: each team gets an adapter and two drawers — how this team does it, what this team knows.

Get involved

Questions, reproductions that disagree with a number here, and proposals are welcome as issues: lora-kernel · agentvcs · gene-evidence · everything else under github.com/EvolvingAgentsLabs.