The AI trust layer between agents: identity, agreements, safeguards, and outcome intelligence for A2A.

Viewing files at commit 68c3fd2f8af0Commit details View latest

docs/ARCHITECTURE.md 16 lines · 1.1 KB

Architecture

OpenClasp surrounds existing agent transport; it does not carry or rewrite ordinary conversation content.

Requester agent -> requester sidecar -> existing A2A transport -> provider sidecar -> provider agent
                         |                                      |
                  local policy/events                    local policy/events
                         +---- permitted structured records ----+
                                         |
                              OpenClasp local/network API

The protocol package owns wire validation and cryptography. The core package is transport-independent and keeps deterministic authorization separate from suggestions. REST, SDK, MCP, sidecar, CLI, and dashboard call the same core behavior. AuditStore separates persistence; the MVP supplies memory and SQLite implementations and can later add PostgreSQL without changing policy logic.

Deterministic failures return DENY. Evidence or behavioural uncertainty returns CHALLENGE. ALLOW remains contextual to task, authority, data, version, and evidence.

# Architecture

OpenClasp surrounds existing agent transport; it does not carry or rewrite ordinary conversation content.

```text
Requester agent -> requester sidecar -> existing A2A transport -> provider sidecar -> provider agent
                         |                                      |
                  local policy/events                    local policy/events
                         +---- permitted structured records ----+
                                         |
                              OpenClasp local/network API
```

The protocol package owns wire validation and cryptography. The core package is transport-independent and keeps deterministic authorization separate from suggestions. REST, SDK, MCP, sidecar, CLI, and dashboard call the same core behavior. `AuditStore` separates persistence; the MVP supplies memory and SQLite implementations and can later add PostgreSQL without changing policy logic.

Deterministic failures return `DENY`. Evidence or behavioural uncertainty returns `CHALLENGE`. `ALLOW` remains contextual to task, authority, data, version, and evidence.

1	`# Architecture`
2
3	`OpenClasp surrounds existing agent transport; it does not carry or rewrite ordinary conversation content.`
4
5	```text
6	`Requester agent -> requester sidecar -> existing A2A transport -> provider sidecar -> provider agent`
7	`\| \|`
8	`local policy/events local policy/events`
9	`+---- permitted structured records ----+`
10	`\|`
11	`OpenClasp local/network API`
12	```
13
14	The protocol package owns wire validation and cryptography. The core package is transport-independent and keeps deterministic authorization separate from suggestions. REST, SDK, MCP, sidecar, CLI, and dashboard call the same core behavior. `AuditStore` separates persistence; the MVP supplies memory and SQLite implementations and can later add PostgreSQL without changing policy logic.
15
16	Deterministic failures return `DENY`. Evidence or behavioural uncertainty returns `CHALLENGE`. `ALLOW` remains contextual to task, authority, data, version, and evidence.