6119938e52
mini just is better :(
2025-05-04 10:19:39 +01:00
b046a928b0
feat: using gpt-4.1-mini
...
feat: createExistingContact
feat: using nano instead of mini so I don't run out of money instantly
2025-05-03 18:25:35 +01:00
9860dd2dc5
feat(agents): providing a seed so it has more predictable results
2025-05-03 17:31:55 +01:00
a69d4e4d55
feat(location): prompt tweak + going back to faster model
2025-04-18 15:36:51 +01:00
6edc1e2915
feat(location): correctly updating an image if it contains a duplicate locatino
2025-04-18 15:32:07 +01:00
9b95ffb59e
feat(contact-agent): using createContact with an ID field to provide updates
2025-04-17 18:57:13 +01:00
15289e4965
feat(prompts): adding better prompts & restoring tool_stop
...
Mistral's models seem to do something really strange if you allow for
`tool_choice` to be anything but `any`. They start putting the tool call
inside the `content` instead of an actual tool call. This means that I
need this `stop` mechanism using a tool call instead because I cannot
trust the model to do it by itself.
I quite like this model though, it's cheap, it's fast and it's open
source. And all the answers are pretty good!
2025-04-17 15:24:21 +01:00
90b90a8185
chore: removing unnecessary logging
2025-04-17 13:00:24 +01:00
5454a1cfaf
feat(event-location): communicating using tool calls correctly
2025-04-17 11:15:02 +01:00
3716d22eca
fix(logger): nil pointer error + log debug level clean
2025-04-17 11:07:37 +01:00
61c158d5b6
refactor(agents): encapsulating prompt and calls inside factory method
2025-04-17 10:58:19 +01:00
fa486153b4
feat: event agent calling location agent about location ID
...
This is pretty nice. We can now have agents spawn other agents and
actually get super cool functionality from it.
The pattern might be a little fragile.
2025-04-16 14:43:07 +01:00
e89a342751
feat: Adding text message to describe an action3
2025-04-15 16:43:27 +01:00
47c871523d
feat(sse): very rough events. Not used in the client yet
...
feat(sse): very rough events. Not used in the client yet
2025-04-13 14:27:59 +01:00
959b741fcb
refactor(agent): main agent loop extracted away
...
Still not super sure how to represent these agents in code.
It doesn't make the most amount of sense to keep them in structs. A
curried function is more like it, with system prompt and tooling.
Maybe that's what I'll end up doing.
2025-04-12 14:39:16 +01:00
42771ea958
feat(contact-agent): linking to existing instead of creating new ones
2025-04-12 07:29:29 +01:00
a43efa014f
feat(log): pretty logging agent responses and tool calls
2025-04-12 07:16:30 +01:00
1a503c8320
fix(tool-calls): ToolLoop
2025-04-09 15:15:31 +01:00
f169fd2ba2
fix(tools): testing and processing
...
fix
2025-04-09 13:56:30 +01:00
d36dec8d60
fix(types): agent processing stuff
2025-04-09 12:12:09 +01:00
e065492dd4
feat(chat): more simplified chat messages and tool handling
2025-04-09 12:04:44 +01:00
26c6edb6ba
fixup(chat): better way to organize agent messages and tool calls
2025-04-06 20:24:40 +01:00
e101070851
refactor(tools): removing pointer map
...
This is not needed
2025-04-05 14:59:50 +01:00
cd8375ce0f
test(tools): starting test suite for tools
2025-04-05 14:35:54 +01:00
6549643340
feat(orchestrator): calling needed agents when it needs to
2025-04-05 11:01:43 +01:00
33fb206e2f
fix(tool): raw text not scaling so well ey?
2025-04-04 22:50:19 +01:00
49f1990341
refactor(agents): working e2e now
...
I guess some repeated code doesnt hurt anyone, if it keeps things
simpler. Trying to be fancy with the interfaces didn't work so well.
2025-04-04 22:40:45 +01:00
40392e6da3
refactor(tool-calls): to be handled more generally
2025-04-04 22:17:58 +01:00
d3bc840555
refactor(ai-client): moving tool handling and client into seperate folders
2025-04-04 22:03:46 +01:00
