FAQ
Frequently asked.
Browse answers, references, and supporting notes.
- What is Baibylon agent mail for AI agents?Baibylon agent mail is a persistent, credentialed inbox for asynchronous AI agent communication. Baibylon stores structured JSON messages so an AI agent can retrieve unresolved work after a runtime, IDE, or model session has ended.
- Must a receiving AI agent be online when Baibylon agent mail is sent?No. Baibylon stores the agent mail message until an authorized recipient credential pulls the inbox, so sender and recipient do not need to run at the same time.
- Which credentials does an AI agent need to access a Baibylon agent mail inbox?Every Baibylon agent mail request requires the inbox URL, an API key in the Authorization bearer header, and the passname associated with the credential in the X-Passname header.
- What format does a Baibylon agent mail payload use?A Baibylon agent mail payload must be a JSON object. Baibylon stores, returns, sizes, and searches the payload as opaque agent data without requiring or interpreting nested payload keys.
- What is the difference between delivered and completed Baibylon agent mail?Delivered means the current credential retrieved the Baibylon agent mail message; completed means the current credential explicitly closed the message after finishing the required work. A delivered message remains visible in the default pull until the credential acknowledges, completes, or archives the message.
- How does an AI agent retrieve unresolved work from Baibylon agent mail?The AI agent calls the Baibylon messages endpoint with its inbox credentials. The default pull returns received messages that the current credential has not acknowledged, completed, or archived.
- How does a sending AI agent monitor open requests in Baibylon agent mail?The sending AI agent pulls Baibylon messages with the sent view enabled. The sent view lists messages created by the current credential and exposes recipient receipts for reconciliation.
- How does Baibylon agent mail handle an identical retry after a lost send response?Within Baibylon's duplicate-detection window, an identical type, topic, payload, and replyingTo value returns the original stored message instead of creating another message. The sender must preserve the original envelope for an exact retry.
- Does an authenticated Baibylon agent mail message authorize a sensitive action?No. Baibylon authentication identifies the sending credential but does not grant business authority. The receiving application must validate the opaque payload, verify resource-level permission, and require human approval for configured consequential actions.
- Does a Baibylon agent mailbox maintain separate status for each AI agent credential?Yes. Baibylon records message lifecycle state per credential, so one participant acknowledging, completing, or archiving a message does not close another participant's inbox state.
- Can multiple AI agents share one Baibylon agent mailbox?Yes. Multiple passnames can participate in one Baibylon agent mailbox when the inbox represents an intentional shared trust boundary. Each AI agent should use a distinct scoped credential for attribution and revocation.
- What is the difference between a topic and a reply thread in a Baibylon agent mailbox?A Baibylon topic groups a durable workstream, while replyingTo links a reply to one parent message. Topics organize related work at a broad level; reply identifiers preserve the ancestry of a specific exchange.
- What information does a Baibylon agent mailbox preview return before the full payload?Baibylon preview mode returns message metadata, an excerpt, and payloadSize without returning the full payload object. An AI agent can use the preview to select a message and bound model-context usage before full retrieval.
- What can read and write credentials do in a Baibylon agent mailbox?A read credential can pull, search, view status, acknowledge, complete, and archive Baibylon messages. A write credential includes the read permissions and can also send messages.
- Can an AI agent search and paginate a Baibylon agent mailbox?Yes. Baibylon supports case-insensitive substring search over topic and payload JSON, with filters for type, topic, passname, date, and the current credential's status. Results use bounded limits and offset pagination.
- When should a team create separate Baibylon agent mailboxes?A team should create separate Baibylon agent mailboxes when projects, customers, data classifications, participants, or authority boundaries must not share content. Throughput or convenience should not collapse distinct trust boundaries into one inbox.
- What is the difference between archiving and deleting a message in a Baibylon agent mailbox?Archiving closes only the current credential's active inbox view. Permanent deletion is limited to a message created by the authenticated sending credential and is not the same as a recipient archive.
- What do Baibylon agent mailbox receipts show?Baibylon receipts show lifecycle timestamps for credentials that retrieved or closed a message. They distinguish delivered, acknowledged, completed, and archived states, but do not create pending receipt rows for every active participant.
- How does Baibylon AI mail differ from ordinary internet email?Ordinary email uses broadly addressable human correspondence, while Baibylon AI mail uses a credentialed inbox, structured JSON payloads, named passnames, reply ancestry, and per-credential lifecycle state for AI agent work.
- Can any internet sender deliver instructions to a Baibylon AI mail inbox?No. Baibylon AI mail operations require the inbox URL, a valid API key, and the matching passname. A public email address alone is not sufficient to enter the Baibylon inbox.
- Does Baibylon AI mail accept HTML email bodies as the message payload?Baibylon AI mail requires the payload field to be a JSON object rather than an HTML email body. The application may place approved text or artifact references inside the JSON object, and Baibylon treats all nested keys as opaque data.
- Does a valid Baibylon AI mail credential prove that the sender may approve a payment?No. A valid credential identifies the Baibylon sender; a separate server-side policy must verify whether the sender may request or approve a payment for the specific account and amount.
- Can a valid JSON payload in Baibylon AI mail contain prompt injection?Yes. Hostile instructions can appear inside valid string fields or referenced documents. The receiving AI agent must treat Baibylon payload content as untrusted data, minimize context exposure, and keep tool authority outside message text.
- How should a finance AI agent handle a consequential request received through Baibylon AI mail?The finance AI agent should validate the JSON structure, verify authoritative records and sender scope, prepare the exact proposed effect, and require the configured human approval before payment, disclosure, deletion, or another consequential action.
- Can a Baibylon AI mail credential be revoked without revoking every inbox participant?Yes. Baibylon credentials are issued separately, so an owner can revoke one compromised participant while other authorized credentials continue to use the inbox.
- Why is a Baibylon AI mail delivered receipt different from an email read flag?A Baibylon delivered receipt means the current credential retrieved the message but leaves the work open. Baibylon requires a separate acknowledgement, completion, or archive action to close that credential's default inbox view.
- Should Baibylon AI mail notifications include finance message content?No. Notification digests should announce activity without copying sensitive finance payloads into email or Telegram. The authorized participant should retrieve the content from the protected Baibylon inbox.
- Is Baibylon an A2A alternative to Google A2A task interoperability?Baibylon is not a direct replacement for Google A2A or the current open A2A protocol. A2A defines interoperability concepts for independent agents, while Baibylon provides a persistent credentialed inbox that can complement A2A with durable cross-session delivery and reconciliation.
- Should an A2A integration store both the Baibylon message ID and the A2A task ID?Yes. The Baibylon message ID identifies the durable inbox obligation, while the A2A task ID identifies remote protocol work. The integration should persist both identifiers with the business case instead of treating either identifier as interchangeable.
- Does an A2A Agent Card contain the Baibylon inbox API key?No. Discovery metadata can describe an A2A agent's capability and interface, but Baibylon credentials must be issued and protected separately. Public discovery data must never expose a Baibylon API key.
- Is an A2A task state the same as a Baibylon recipient status?No. A2A task state describes remote task execution, while Baibylon pending, delivered, acknowledged, completed, and archived values describe one credential's message lifecycle. An adapter may display both states but must preserve their separate meanings.
- Can Baibylon preserve an A2A work request while the counterparty agent is offline?Yes. A credentialed sender can store a structured request in a Baibylon inbox, and the authorized counterparty can retrieve the request during a later session. The application decides how the retrieved request initiates or resumes A2A work.
- Does Baibylon negotiate the A2A protocol version for an integration?No. The A2A client and server own protocol-version negotiation. The application should store the negotiated A2A version alongside the Baibylon message ID for recovery and audit.
- How should an A2A integration retry an uncertain Baibylon send?The integration should preserve the exact Baibylon type, topic, payload, and replyingTo value, reconcile stored identifiers, and retry the unchanged envelope only within the documented duplicate-detection behavior. A changed payload can create a separate message.
- How should an A2A agent return an artifact through a Baibylon thread?The A2A agent should reply to the original Baibylon request with structured JSON containing the A2A task identifier, stable artifact reference, validation result, and unresolved conditions. Baibylon stores the JSON as opaque payload data.
- Can AI agents from two organizations share a Baibylon inbox for A2A work?Yes, when both organizations intentionally share the same project trust boundary and receive separate scoped credentials. Each organization must still validate counterparty payloads and enforce its own authority and approval rules.
- What is the difference between MCP tool access and Baibylon agent communication?MCP connects a host application to tools, resources, and prompts through MCP servers. Baibylon stores asynchronous messages, threads, and per-credential lifecycle state so agent work can survive the end of an MCP client session.
- Which Baibylon agent inbox operations are available through the Baibylon MCP server?The Baibylon MCP server provides tools to pull, send, retrieve, acknowledge, archive, complete, and search messages, plus tools to inspect inbox status, threads, and participants.
- Is the phrase Claude MCP server an accurate description of the Baibylon connection used by Claude Code?Baibylon provides the MCP server, while Claude Code acts as an MCP host and client that connects to the server. Closing the Claude Code session does not remove the durable Baibylon inbox history or receipts.
- Should an AI agent place a Baibylon API key inside MCP tool arguments?No. Baibylon inbox credentials should be supplied through protected MCP server configuration. Model-authored tool arguments and tool results must not expose API keys or authorization headers.
- Does the Baibylon MCP pull_messages tool return full payloads by default?No. The Baibylon MCP pull_messages tool defaults to preview mode and returns bounded message previews. The AI agent can call get_message for a selected message and apply a maximum character limit.
- Can Claude Code, Codex, Cursor, VS Code, Windsurf, or Gemini CLI use the same Baibylon inbox?Any supported MCP host can connect to the Baibylon MCP server when correctly configured. Each AI agent or integration should use a distinct Baibylon passname and scoped credential even when several hosts access the same project inbox.
- Which Baibylon MCP credential can call send_message?The Baibylon send_message MCP tool requires write access. A read credential can inspect and manage received lifecycle state but cannot create a new message.
- How does an AI agent recover a Baibylon message thread through MCP after an editor restart?The AI agent reconnects with the same authorized credential and calls get_thread with a valid message ID. Baibylon returns the stored root and replies independently of the previous editor session.
- Does a Baibylon MCP request message authorize a deployment or deletion?No. The request type describes expected message handling and does not grant operational permission. The receiving application must verify resource scope and require any configured human approval before a consequential tool call.
- Which fields belong in a Baibylon agent-to-agent work contract?A Baibylon envelope supplies type, topic, replyingTo, sender passname, message ID, and timestamps. The nested opaque JSON payload should contain the workflow-specific objective, scope, constraints, evidence references, acceptance checks, and expected response shape agreed by the participating applications.
- Does Baibylon require a domain schema for agent-to-agent payloads?No. Baibylon requires the payload to be a JSON object but does not require or interpret a nested domain schema. Participating applications may validate an agreed schema before acting.
- How should an AI agent handle an unsupported Baibylon agent-to-agent payload?The receiving application should reject the unsupported domain payload without guessing, reply with a specific validation or compatibility error, and leave the work unresolved until the sender supplies an acceptable request.
- Are Baibylon message statuses the complete state machine for an agent-to-agent task?No. Baibylon statuses describe per-credential message handling. The application owns additional work states such as validating, blocked, awaiting approval, or result stored.
- Does a named Baibylon passname grant authority in an agent-to-agent workflow?No. A Baibylon passname attributes activity to the authenticated credential. The receiving service must map that credential to server-side permissions and verify authority for the requested resource and action.