Independent AI products & research · Germany
IlhamLabs designs and ships tools for multi-model collaboration, private AI, and reliable agent workflows for product teams, researchers, and technical partners.
Evidence, not adjectives
Published Praxium on PyPI Licensed Apache 2.0 tooling Documented Public architectureSelected work
Each flagship starts with a specific constraint: too many disconnected model chats, sensitive data leaving a device, or messages that should arrive at the right moment.
A graph-native workspace for branching conversations, comparing models, and merging useful results without losing the thinking that led there.
A mobile AI workspace designed to keep conversations local: no account, no cloud chat history, and no required model lock-in.
A deliberate place to write, encrypt, and schedule messages that become available only at the chosen future moment.
A calm AI companion experience with distinct voices, persistent context, personal controls, and clear disclosure that the companion is not human.
More products
Collaborative study sessions with Pomodoro timing, chat, and personal notes.
View project ↗Receipt scanning, spending analytics, and practical financial insights.
View project ↗Ingredient analysis that helps users investigate whether a product may be Halal.
View project ↗Research & infrastructure
Published tools are linked to their code, documentation, license, and package. Earlier research directions are labeled clearly as work in progress.
A typed, asynchronous Python framework for AI agents and graph workflows. It supports deterministic routing, structured events, retries, timeouts, cancellation, checkpoints, and provider-neutral model adapters.
# Install the alpha
$ pip install praxium
from praxium import GraphBuilder, Runtime
graph = (
GraphBuilder("research-flow")
.add_node("observe", observe)
.add_node("evaluate", evaluate)
.set_entrypoint("observe")
.build()
)
result = await Runtime().run(graph, data)Investigating how simulated states such as curiosity, boredom, and excitement influence exploration and decision-making in autonomous agents.
Designing comparable training runs for supervised, REINFORCE, and PPO approaches with explicit reward functions.
Responsible by practice
“Responsible AI” only means something when it changes product decisions. These principles connect directly to shipped interfaces and public technical work.
Users should know when they are interacting with AI, what stage a product is in, and where its limits are.
See it in Amorae ↗When a useful experience can run locally, privacy should be an architectural property, not just a promise.
See it in ilmora ↗Structured events, deterministic test providers, and documented workflows make agent behavior easier to inspect.
Read the architecture ↗Alpha software, early-access products, and research directions are labeled before visitors decide to use them.
See the alpha status ↗About IlhamLabs
IlhamLabs is an independent AI product and research lab based in Germany. The work combines product development with applied research: build something useful, document the difficult parts, and turn repeatable foundations into shared infrastructure.
Collaboration stays direct and technical. IlhamLabs works best with product teams, researchers, and organizations exploring multi-model workflows, privacy-first AI, or agent prototypes.
Turn a defined AI use case into a focused, testable product experience.
Design provider-neutral orchestration, tool use, routing, and evaluation flows.
Explore on-device or controlled-deployment approaches for sensitive use cases.
Current direction
Updated August 2026
Public package, documentation, tests, architecture, and Apache 2.0 license.
Early-access learning for Amorae and continued refinement across flagship products.
Clearer public reporting for autonomous-agent experiments and model comparisons.
Start a conversation
Share the problem, the people it affects, and what a useful first outcome would look like. IlhamLabs aims to reply within two business days.