Aider already loads read-only context files. Give it one that stops overpaying.

How this maps onto Aider

No auto-discovery here — you load the file, once, and it sticks.

Aider's own docs are explicit that it has no built-in rules-file or Agent Skills directory it scans on its own — per Aider's conventions docs, the documented pattern is a conventionally-named CONVENTIONS.md you load explicitly, either with aider --read CONVENTIONS.md on the command line or /read CONVENTIONS.md inside a running chat. Either way marks the file read-only and, per that same page, caches it if prompt caching is enabled — cheaper on every subsequent turn, not just the first.

To avoid typing --read every session, Aider's config docs describe a persistent alternative: a read: entry in .aider.conf.yml, which accepts one file or a list. Aider looks for that config file in your home directory, your git repo root, and your current directory, loading all three if present — later ones override earlier ones — or you can point it at a specific file with --config/-c. Once read: names the policy file, every session in that repo loads it automatically, no flag required.

Separately, the agents.md project's own compatibility page lists Aider as reading AGENTS.md when you add read: AGENTS.md to .aider.conf.yml — functionally the same read: mechanism above, just pointed at the cross-tool convention instead of a bespoke file. That claim comes from the AGENTS.md spec's own site, not from Aider's documentation directly, so treat it as one config line to try rather than a separately-confirmed Aider feature — see Honest limits below.

Aider already has its own cost-tiering primitive worth knowing about: --model, --weak-model, and --editor-model assign different models to fixed roles — main chat, commit messages/history summarization, and editing, respectively (per Aider's options reference). That's a static, role-based split decided once at launch. Undercut's tiers — cheap, standard, frontier, apex — are dynamic instead: assigned per unit of work based on verifiability and risk, escalating only on measured evidence. The two aren't in conflict — --weak-model picks who writes your commit message; Undercut's rubric picks who does the actual edit — but they answer different questions, and nothing here overrides your --model flags or touches Aider's network calls to your LLM provider.

Install for Aider

Free. MIT. Nothing installs, nothing proxies your traffic.

Option 1 — persist it in .aider.conf.yml (recommended, confirmed path)

Run this from your git repo root. It downloads the policy as CONVENTIONS.md — Aider's own documented convention name — and adds a read: line so every future session picks it up with no flag needed.

Option 2 — load it for one session, no config change

Skip the config file and pass --read directly, or run /read CONVENTIONS.md inside a chat you've already started. Either way Aider marks it read-only and caches it if prompt caching is enabled, per Aider's own docs.

Validate it yourself

1. 1Read the published results — testing/README.md and the raw JSON in testing/results/
2. 2Reproduce for free, no API key, no spend: node src/main.js --mock
3. 3Run aider --read CONVENTIONS.md on a small task and check whether it references the routing policy before starting mechanical work

Zero lock-in. It's a markdown file Aider reads — drop the read: line, or stop passing --read, and you're back to status quo.

Six flags assign a tier. Three triggers are the only way up.

This is a named, versioned spec — not internal logic we could quietly change. When the rubric changes (a new flag, a different cap), the version number moves and the changelog says why.

Every unit of work gets scored against six flags, then assigned the cheapest tier that can plausibly handle it. It only moves up a tier when something objective proves it needs to — a failed check, a measured disagreement, an explicit uncertainty flag. It never moves up on a guess, and it never moves back down mid-task.

 unit of work
      |
  6-flag rubric
      |
  +---+----+
  | 0 flags|--> cheap
  | 1-2    |--> standard
  | 3+/own |--> frontier
  +---+----+
      | fail x2 / disagree / uncertain
      v
  escalate +1 tier
  (residue only)
      |
  still unresolved?
      v
  single batched
  apex tie-break

Why it holds up

Flags steer. Verification and escalation decide.

Even when the rubric flags are scored imperfectly, the escalator still routes work to the right tier almost every time — because a wrong flag only ever costs one extra cheap attempt, never a wrong answer or a big bill.

Honest limits

What this doesn't do.

Not a proxy, not a gateway, doesn't auto-flag in production — Aider scores the flags itself (measured robust to getting them wrong, see above). Unlike most clients on this site, Aider won't pick the policy up on its own: forget the read: line or the --read flag and it's simply not loaded, silently. The AGENTS.md path above is a third-party claim from the agents.md project's own site, not Aider's documentation, so treat it as unverified until confirmed directly. Doesn't promise a dollar figure for your workload, and doesn't prove your number until you run it — every figure here is a reproducible observation from our tasks and graders, not a guarantee about yours. Full list of general limits, with the finance/procurement note and the paid-tier disclosure, is on the main page.