Agents orchestrate. Tools execute. The Ledger remembers.
EveryShall is built as a mesh of narrow agents rather than one assistant. Each agent has a single responsibility, a fixed set of tools it may call, and an obligation to write what it did to the Ledger.
Six agents
| Agent | Responsibility | Deterministic tools |
|---|---|---|
| Scout | Ingests new solicitations from SAM.gov and the state and local portals you configure, deduplicates them, and downloads every attachment. | API pulls, feed reads, document download, deduplication |
| Fit | Scores each solicitation against your go/no-go rules, NAICS codes, set-asides, and past performance, and writes out the reasons for the score. | Rule evaluation, retrieval over your firm documents |
| Compliance | Reads the full solicitation package and builds a matrix of every "shall", "must", deadline, format rule, and required form. | Document extraction, structured table build |
| Drafter | Produces an outline and first-draft sections. Every sentence is tied to one of your documents or to the solicitation, or it is flagged as unsupported. | Retrieval, generation with citations |
| Reviewer | Checks the draft against the compliance matrix and lists gaps, unsupported claims, and page-limit problems before you see it. | Rule checks against the matrix |
| Ledger | Records every decision, source, model call, and human edit in an append-only log you can open for any bid. | Append-only audit log |
Reasoning and execution are separated on purpose.
Models decide what to do next
A model classifies a candidate sentence as a staffing requirement. It does not decide which sentences are candidates, and it does not decide what the requirement says: candidates are selected lexically, and the verbatim text is byte-checked against its source span before the record is accepted.
Tools do the work
Source polling, document normalization, rule evaluation, retrieval, table construction, and file export are ordinary code with ordinary tests. Each tool call is logged with its arguments and a hash of its result. When a tool fails, the step fails and the Ledger shows exactly where.
Every sentence carries a source
Drafter writes only from retrieved passages and attaches, per sentence, the (document hash, page, character range) it used. Reviewer re-verifies each sentence in a separate call that sees only the sentence and its spans. Unsupported sentences are flagged in the editor and blocked from export until acknowledged.
Humans hold the pen at the end
No agent has a tool that writes to any external system, and no portal credentials are stored. You review, edit, export, and submit. Every edit is a Ledger entry with the text before and after.
An append-only record for every bid
For each solicitation you pursue, the Ledger holds the ordered list of what happened. It is the answer to "why did it say that?" and to "who changed this?"
| Entry | Recorded |
|---|---|
| Ingestion | Source portal, notice ID, timestamp, attachments fetched and their hashes |
| Fit decision | Score, the rules that fired, the documents retrieved, the model and prompt version used |
| Compliance matrix | Each requirement with its verbatim text and page reference in the solicitation |
| Draft sentence | Text, cited source document and location, or an "unsupported" flag |
| Review | Gaps found, page counts, unsupported sentences listed |
| Human edit | Who changed what, when, and the text before and after |
| Export | File produced, format, timestamp, and the matrix version it was checked against |
What is connected today
This table is kept current. A connector appears here only after it has been run against live postings.
| System | Direction | Status |
|---|---|---|
| SAM.gov Contract Opportunities API | Read | In build |
| State and local procurement portals | Read, one connector per portal | Planned |
| Document upload (PDF, Word, text) | Read | In build |
| Export to .docx and .pdf | Write, to you | In build |
| Email digests and alerts | Write, to you | In build |
| Any procurement portal submission | None, by design | Never |
Hosted in the United States. Built to be inspected.
Models
Two model tiers: a small model for classification and scoring, a large model for extraction, drafting, and verification. Model identifier and prompt version are recorded on every call. No customer data is used for training.
Orchestration
Each solicitation runs through the agents as a retryable, step-by-step workflow whose execution history is part of the Ledger.
Storage and access
Encrypted storage, separated per firm, with sign-in protected by multi-factor authentication. Details on the Security page.
We are onboarding a small number of pilot firms.
Bring a live solicitation to the first call. We will build its compliance matrix while you watch.