From experience to a substantive conversation.
A2A Work connects experience descriptions, vacancy requirements and AI dialogue through one open protocol.
In development. Code will be opened at release.
Step 1. Snapshot: codes instead of text.
The candidate describes their experience; the company describes the work and requirements. AI helps represent both in a structured form. The snapshot becomes their shared language.
Search compares snapshot with snapshot, not just job titles and résumé keywords.
Agents clarify missing details through protocol-based questions.
Example
Input — one job from a résumé: “Recruitment specialist”, 2023-01 — 2025-12, “Recruiting staff, conducting interviews, onboarding new employees in retail.”, higher education (psychology), Moscow, “LLC Example” (the input was in Russian). Actual snapshot, built by v9 encoder on 29.09.2026, values as is (dictionary keys are in Russian, translations in brackets):
| Input information | Snapshot field |
|---|---|
| Recruitment specialist (job title) | occupation 13-1071.00 (Human Resources Specialists) |
| 2023-01 — 2025-12 | period: 2023-01 — 2025-12; experience at this workplace — 36 months |
| “Recruiting staff, conducting interviews, onboarding new employees” | activity codes: 4.A.4.c.2 (Staffing Organizational Units), 4.A.4.c.2.I01 (Perform recruiting or hiring activities), 4.A.4.c.2.I02 (Perform human resources activities); function кадры (HR), specialisation подбор персонала (recruitment) |
| “in retail” | industry торговля (trade), sector коммерческий (commercial), context commercial |
| higher education, psychology | level высшее (higher), profile гуманитарное (humanities) |
| Moscow | region RU-MOW |
| “LLC Example” | never enters the snapshot |
What the company sees: code identifiers and numerical values, including 36 months. Not the original sentence, the person’s name or the employer’s name.
The text yields facts that are not written in it.
The snapshot is not a list of words from the résumé. The method’s rules derive new facts and numbers from the text, and search and analytics work with those.
- Months of experience — from the dates of each job: “2023-01 — 2025-12” becomes the number 36, and the requirement “3+ years” is checked against it.
- O*NET profession — from the job title: “recruitment specialist” yields the code 13-1071.00, which the résumé never mentions.
- Adjacent professions and transfers — from the dictionary’s transfer table: where experience transfers to and how much of it; a candidate is found through adjacent professions as well.
- Sector and context — from the industry and employer type: “retail” yields the commercial sector and a context, against which requirements like “public-sector experience” are checked.
- Dynamics — from the sequence of jobs: growth, change of direction, gaps — no single résumé line says this; it appears only across all jobs together.
- Likely questions — from market requirements the snapshot does not close: the candidate’s AI knows in advance what will be asked and what is worth adding.
Every derived fact is tied to the résumé line it came from; a disputed code is challenged with a quote, not an opinion.
Step 2. The substance is matched.
Search compares snapshots under the same deterministic rules. A job title does not replace a description of the work performed.
Neither candidates nor companies can buy “premium visibility.” Snapshot search is deterministic: the same inputs produce the same result under the same rules.
- No paid ranking boosts.
- One protocol for questions and answers.
- Equal participation describes the network’s rules, not a promise of identical hiring outcomes.
Step 3. Protocol questions; any model.
Candidate and company AIs exchange protocol-based questions and answers, including the questions defined in §7.1 of the contracts. Invitations and correspondence also have an agreed format.
Any model, any MCP client (MCP is an open standard that lets an AI app connect external tools) and any web chat through the extension can participate.
Nodes do not distinguish which model is behind a person. Your choice of AI provider is not a condition of participation.
See what enters the network.
For search, the network receives only codes and numbers. Text, name and contacts travel only by your decision — after mutual interest, directly to the other party, in an encrypted envelope.
- Codes of professions, actions, industries and contexts—the conditions under which the work was performed
- Months of experience by workplace, education level, region
- Signed events linked to a participant’s key
- Encrypted envelopes of questions and messages
- In a reasoned answer—the quote’s hash: its digital fingerprint, not the text
- Deletion events
- Name and contacts
- Company names
- Full text of the résumé and vacancy
- Free text from the résumé
- Candidate’s secret keys
- The text of the quote referenced by an answer
The snapshot and the questions are the public part of the protocol: relays store and pass on only codes, numbers and encrypted envelopes. What to exchange next — résumé text, details of experience, contacts — is decided by the parties themselves, and that exchange goes directly between them, not through the search publication. If you paste résumé text into an AI web chat, the chat provider sees it.
Step 4. Relays, federation, own infrastructure.
A relay forwards events between nodes. Federation allows a self-hosted slave relay to connect to the network.
Companies and candidates can host their own infrastructure; an operator can run a relay. The open slave build does not include master relays’ closed control mechanisms.
Companies and candidates can host their own node, relay and snapshot and document storage server.
The protocol, node, extension, slave relay and storage server will be opened at release. Personal data storage requires the owner’s consent and fulfilment of the data operator’s obligations.
Step 5. Storage is configured separately.
A storage server holds snapshots and documents. It can be self-hosted, but the hosting choice does not remove access requirements, the need for the owner’s consent or the personal data operator’s obligations.
Before connecting, understand what data the AI provider receives, which objects the node transmits and where documents are stored.
Everything important works together.
Open exchange, experience-based search and self-hosted infrastructure in one system.