Agent Abilities

Mobility

• move from machine to machine.

Autonomous

• each agent is its own thread
can find its own resources

Cross platform

• interpreted to some extent...maybe Java, maybe XML.

I was thinking about this and it might be interesting (although very difficult) to completely specify the agent in XML...then the agent could be compressed and encrypted easily.

Secure

• support multiple types of encryption
• multiple authentication policies

Currently my thinking on this is that we have an option for security, without it the system does no authentication. With it, SSL is used and before accepting an agent the connecting server must be authenticated in some way. Probably with a public key/private key exchange.

Although, we should be able to support mutiple authentication methods...so when a server contacts another, the first thing that happens is that the server announces what types of authentication it supports. Then the connecting server can decide which to use, depending on what support it may have available.

• optionally encrypted user data.

There is large concern that a user's profile (carried by an agent) can be compromised and email addresses, credit card numbers and other sensitive information be revealed unwittingly. We'll have to allow an option to encrypt all data carried by the agent, so that only authorized data may be accessed.

We should also apply good accounting practices, keeping track of who accesses what personal data, etc.

• secure inter-agent authentication/communication.

Before releasing any data, authentication needs to occur if we allow any agent to agent communication.

Interactive

• capable of 'finding' other agents
• capable of agent-to-agent interaction

Error handling

• survive across reboots
• intelligent network traversal

Multiple decision support

• pluggable intelligence
• support for bayesian nets
• support for neural networks
• support for finite state machines

System Abilities

Bootstrap server

• The 'main' of the agent system. This should be fairly lightweight, starting only what is specified in the config file. Possibly not even starting an agent listener unless it is specified. In this case the server would act like a more traditional server, serving information on agent and server whereabouts. Or...maybe if a port number is not specified it doesn't even open a port but allows agents and servers to run locally.
• Everything that is specified in the config file should implement Runnable and be started in its own thread. If a class in the config file is a server, it is handed off to the server factory, if it's an agent it's handed off to the agent factory and if it's anything else the 'run' method is called.

Agent Factory

• can create templated agents. A server or agent may request that an agent be created. The agent factory may also handle the creation of agents at startup.

Server Factory

• allows for servers to be created in a templated manner. An agent may arrive at a machine and request that a server be created by the server factory. The server factory may also handle any servers at startup.

the servers can be registered for...by agents or other programs. A garbage collection-like thread could periodically check all the handles to make sure the server is still needed. We should also provide for 'persistent' servers...the ability for an agent or program to request that a server remain available even if the requesting object is no longer present on the server host.

Agent control

• track agents
• start, stop, kill agents
• dynamically reprogram agents, or suggest other programs.
• maybe the system just broadcasts things that need to be done and the agents register for tasks they are suited for.

Agent listing

• the concept of a yellow pages of agents...could help agents find each other.

This could be part of the initial startup sequence, where a current list of active agents is exchanged.

Server Listing

• On startup the agent system reads a config file and locates each of the 'known' agent servers. When connecting to each server a list of 'known' servers is exchanged, strengthening the network. As new servers come on to the network they are announced to every server and added to each server's list of 'known' servers. The list of 'known' servers should be written to a file in case the server goes down and the initial server that was contacted is no longer available.

Security Server

The 'Gatekeeper' for a specific machine. When connected to...this service will announce what forms of authentication are supported, if required. This server will be in charge of authenticating any incoming agents and handling authentication challenges from other servers.

Data Repository

• list common agent data