Layer 01 · Toolkit First release v0.3.0

adgentic

An agentic AI toolkit for business advertising — your agent plans, prices, posts, and learns what sells your services.

Binds to a field of work — business advertising. A toolkit is the domain brain for its field: playbook, engine and guardrails in one repo, handed to any agent.
Get the Forge Pass Back to the explorer
The Benchmark Arena · Shipped proof

Judged in the arena, not in the brochure

The real planner and learning loop, run against three naive baselines in a seeded simulation — 12 simulated months, 8 repetitions, 240 ads per arm. Mean value earned, standard world:

adgentic toolkit+3.97% vs best baseline$23,654.70
population weighting$22,752.28
even spread$19,914.48
uniform placement$19,089.47

The honest summary: a value strategy — fewer conversions than population weighting (45.00 vs 49.00), worth more, because its conversions land in higher-value areas: +3.97% value vs the best baseline in the standard world, +2.02% in the noisy world — and a tie (−0.94%) in the flat control world, where no signal exists and an honest harness should show exactly that. Methodology: docs/BENCHMARKS.md in the repo.

What it is

adgentic, in full

A complete advertising brain for businesses on classifieds, forged into a repo your agent operates end to end. It plans placements by expected profit, prices and positions every ad by neighborhood, runs service campaigns (and product ads for businesses) within platform limits, manages each listing's lifecycle, and learns conservatively from real results.

It never posts, never pays, and never logs in anywhere — plans are files, and execution stays with the operator and their posting tool. The agent does the research, onboarding, planning and learning; the deterministic engine does the math — the same inputs give the same plan, byte for byte, on every rerun.

The strategy core is platform-agnostic: ventures, metric packs, geo adapters and the learning loop know nothing about any one site. Today it plans for Craigslist — the first platform its operator limits and market scanner were researched against; other classifieds arrive through the same adapter pattern, not a rewrite.

Capabilities

Profit-weighted placement planningNeighborhood price & positioning tiersListing lifecycle & price walk-downsPlatform limits & compliance guardrailsMarket scanning & competitor signalsConservative learning from real resultsCampaign planning for services & business offersSelf-updating from the canonical repo

Inside the engine

Math with its reasons showing

102
postal areas scored on 2021 Census data through a 16-signal model, weighted per venture — not gut feel
1.25
premium-boost exponent on the value index in the standard phase — high-value areas pull harder; a 35% cap per municipality keeps plans from over-concentrating
3
listing intents — fast sale, balanced walk-down, or maximum price held patiently; the lifecycle works the one you chose
±10%
the most learning may move a weight in one step — conservative by charter, with every change versioned and explained

Status — honest label

Version 0.3.0 — the first release, shipped

Shipped and working

  • ▸Profit-weighted placement planning with per-venture metric packs deduced from research, not guessed
  • ▸One-off mode — a single sale gets one stateless analysis, no venture created, nothing learned or stored
  • ▸A strategy engine that crosses the pack with current conditions — with all five anti-feedback-loop safeguards as tested mechanics
  • ▸The benchmark arena, one-command demo and visual HTML report — proof machinery shipped in the repo, not promised
  • ▸A limits engine that stores every platform limit with its source and enforces it — caps, cooldowns, pacing, hard stops

Ruled by the guardrails

  • ▸It plans; it never posts or pays. Execution stays with the operator and their posting tool
  • ▸Compliance is machinery, not documentation — limits are config with sources, re-verified against the platform on a standing six-month terms review
  • ▸Learning is conservative — bounded steps, weight floors and ceilings, and it never auto-retires a strategy
  • ▸Your data stays yours — real packs and results live in a local library outside the shipped repo

Keep going

adgentic in the family

agenticlistings

The buy/sell sibling — same platform engine, pointed at casual and professional trading instead of business advertising.

Meet the sibling →

Lingo — built in

Every toolkit ships with Lingo inside: steer adgentic's agent with the same six prefixes you use everywhere else.

See the plugin →

The full explorer

All four toolkits and both plugins, grouped the way the forge files them: by what they bind to.

Browse the catalog →

Product adgentic Layer Toolkit Status First release Access Repo access with the Forge Pass