Rykan V™@rykan_v

Control what your AI can do github.com/christopherkarani…

 United States
rykanv.com
Joined June 2023
765 Photos and videos
Replying to @rykan_v
Link to our telegram channel: t.me/+AuTjaRv5eeQ4MDk0

Bebop

The interstellar Bank 🪐

t.me
3
10
1,967
Rykan V™ retweeted
Christopher Karani@ckarani7
Aug 25
I kept hitting model failures on phones that I couldn't reproduce in the office. LangFuse didn't catch it. LangSmith didn't catch it. So I built the thing I wanted: a flight recorder for AI running on real devices. heavybit.com/library/article…

Why On-Device Inference Needs Custom Observability | Heavybit

Founder Chris Karani explains how hardware state becomes part of the application state when inference goes on-device.

heavybit.com
1
5
178
Rykan V™ retweeted
Christopher Karani@ckarani7
Aug 16
11x faster hooks 13x faster denies 3x smaller binary ryk 0.2.19
1
3
202
it’s a pandemic, your memory is what’s dying
Christopher Karani@ckarani7
Aug 15
@irangareddy Hey — saw your post about the agent wiping the DB / running destructive commands We’ve built Rykan V specifically for this: policy guardrails that block dangerous shell/file actions before they execute. You can also run ryk scan to see every dangerous command your agent has run in past sessions. Open-source here if you want to try or star it: github.com/christopherkarani…
94
Open ryk allowlist for the TUI of what this session can run. Linear dump: ryk allowlist --plain.
33
Run ryk redteam --ci. Same fixtures the CI job uses, same deny path a live session hits.
28
When the session is done, ryk stop. Leaves the policy log on disk so ryk replay still works.
30
After the agent stops, ryk replay --session last. Chat summaries skip the dangerous parts.
22
Rykan V™ retweeted
Christopher Karani@ckarani7
Aug 11
When you run an agent through ryk, we try to put a real OS filesystem boundary around it — Seatbelt on macOS, Landlock on Linux. Not PATH tricks. Kernel rules on the child process. A short walk through how that works.
1
1
62
no one talk about how ai looked at their trash codebase and actually improved it beyond what they could imagine
34
After the agent stops: ryk diff --session last. See staged writes before ryk apply or ryk discard. Review the blast radius, not the chat summary.
44
ryk scan . --secrets before the agent walks the tree. Cheap pass. Catches .env and keys the model will happily cat for you.
37
Pick the mode on purpose: ryk policy check --preset ask | yolo | strict. Same engine. ask prompts, yolo keeps sandbox + fence, strict fails closed.
23
Before you trust a policy, dry-run the scary string: ryk explain "rm -rf /" | ryk test "rm -rf /" --format json. Same decision path the live session uses.
31
Don't assume ryk start left you protected. ryk doctor diagnoses policy, hosts, packs, sandbox backend features, and network policy status. Try: ryk doctor | ryk doctor -v
38
Don't start the agent naked. ryk start creates policy when missing, wires hosts, defaults to Ask on risk, then verifies. Try: ryk start | ryk start --auto --hosts claude,codex
76
Rykan V™ retweeted
Christopher Karani@ckarani7
May 31
Replying to @blader
“it’s really not close” what a load of bs, the model good, Its on par with 5.5 better at following instructions and the honest factor is really underrated. doesn’t write overly defensive code, so less slop. you have a skill issue my friend
11
1
82
28,985
Rykan V™ retweeted
Christopher Karani@ckarani7
Mar 29
Good morning
1
2
107
Rykan V™ retweeted
Christopher Karani@ckarani7
Mar 20
Article

The FileSystem as an Interface for On-Device Agents

I think most people are still talking about agents at the wrong level. The interesting shift is not “models are getting better.” That part is obvious. The interesting shift is that the serious agent

1
1
4
176
Rykan V™ retweeted
Christopher Karani@ckarani7
Mar 19
clone: github.com/christopherkarani… and run this command ./espresso compare --bench --no-power "Hello"

GitHub - christopherkarani/Espresso: Run transformers directly on Apple's Neural Engine in Swift...

Run transformers directly on Apple's Neural Engine in Swift bypass coreml entirely - christopherkarani/Espresso

github.com
1
1
50
Your Swift AI agents just went multiplatform 🚀 SwiftAgents adds Linux support → deploy Agents- to production servers Built on Swift 6.2, running anywhere ⭐️ github.com/christopherkarani…

GitHub - christopherkarani/Swarm: Type-safe tools, on-device inference, multi-agent workflows,...

Type-safe tools, on-device inference, multi-agent workflows, memory, and guardrails, in one Swift-native runtime - christopherkarani/Swarm

github.com
2
6
389
Load more