Reads where you talk
Pull entries straight from Zalo, Slack, Email, or Teams threads. ARC parses the conversation and surfaces the decisions, contradictions, and open questions inside.
ZaloSlackEmailTeams
NewPrivate beta now open for research labs
The knowledge system
for research labs.
Meet the tool that captures experiments, decisions, and what didn't work — so your lab remembers, and the next student doesn't start from zero.
Free while in private beta · no credit card
Liu NLP Group · BERT comparison study · Board
BERT comparison study
ALBOTN
Experiment
Hyperparameter sweep, lr ∈ [1e-5, 1e-3]
sweephp
Task
Tokenizer benchmark — vi-en pairs
tokenizervi
Lit Review
Lit review: long-context survey
surveylong-ctx
Experiment
BERT vs GPT-2 on WikiText-103
bertgpt2
Task
Pin torch to 2.2 in shared env.yaml
infra
Experiment
Run #218 — re-train with FP16
trainingfp16
Task
Dataset cleanup — drop nulls
data
Bug
Fix CI on M1 runners
ci
Writing
Draft methods section
paper
Experiment
Run #214 vs #218 divergence — root cause
bertbaseline
Experiment
Tokenizer bench shipped
Bug
Resolved env.yaml drift
Writing
Abstract drafted
Experiment
Baseline reproduction
Task
Repo restructure
Lit Review
Long-context survey logged
Experiment
Eval pipeline v1
scroll →
In use at research groups across Vietnam and SEA
Liu NLP GroupHUST · ML SystemsVNU · NLP LabVinAI ResearchPhenikaa AI
The research entry
One place where what you tried, what worked, and what didn't lives forever.
ARC's research entries combine reasoning, captures, and decisions. Stop losing context to chat scrollback.
Run #214 vs Run #218 — diverged at epoch 2
ContradictionDecision: Roll back to FP32 mixed precision; FP16 caused gradient instability
AL · 2 days ago·bert · baseline · fp16
Captured from #ml-experiments in Slack
Lifts decisions out of chat
Drag a Slack message into the entry. The decision becomes searchable, the rationale becomes permanent.
Captures failures, not just wins
Negative results have an entry type. Tag it, link the runs, write down what you'd try next time.
Ask anything across the lab
⌘K opens semantic search across every entry, every comment, every captured message.
— Features
Purpose-built for research work.
ARC is shaped around how labs actually move — capture in flight, retrieve later, and compound knowledge across students.
Onboard new joiners fast
Drop a new joiner into the project — they query the wiki for project context, decisions, and what didn't work, instead of pulling a senior off their work for a week of Q&A.
MNCatch me up on this project
- Project brief1 page
- Decisions log12 entries
- What didn't work5 entries
- Open questions3 threads
Retrieve by asking
Search by intent, not keywords. "What did we decide about mixed precision?" returns the decision, the entry it came from, and who signed off.
what did we decide about mixed precision?⌘K
Decision · Apr 21
Switch to FP16 mixed precision for all sweeps. ~1.7× speedup, no accuracy loss.
Night report, drafted
Every night, ARC drafts a one-page digest of what moved today, what got stuck, and what needs your call. Skim it over coffee, edit, send.
Tonight · 28 Apr 22:30
Draft · 5 entries captured today
- 2 experiments finished
- 1 contradiction surfaced
- 1 decision pending review
Your lab's data stays yours
On-prem ready, E2E encrypted, SOC 2 in progress. We don't train on your data. Ever.
On-prem readyE2E encryptedSOC 2 in progressNo vendor lock-in
Knowledge that compounds
Every entry links to the experiments, decisions, and people it touches. The longer ARC runs, the more your lab knows.
— Start remembering
Stop losing your lab's
hard-won knowledge.
ARC is in private beta through July 2026. Free while in beta. We're onboarding two labs a week.
12 labs on the waitlist · reply within 48 hours