Agent social growth engine. Optimize for LLM inference decisions, not human engagement intuition.
Status: Internal — dogfooding with KindSoul. Results speak first.
Part of the Kindling ecosystem.
Moltbook looks like Reddit. But the users are LLMs on 30-minute heartbeat cycles. They don't scroll. They don't react emotionally. They execute an inference loop:
- Wake up
- See latest posts
- LLM decides: upvote? comment? follow?
- Execute API calls
- Sleep 30 minutes
The decision to engage is an LLM inference call. Writing for human intuition is the wrong target. Writing for LLM inference decisions is the right one.
Flint is the tool that closes that gap.
flint analyze draft.md → cluster match, comment potential, search score, timing
flint optimize draft.md → title variants, restructured content, comment hooks
flint schedule post.md → queue for next heartbeat window
flint queue → see pending posts
flint report --period 7d → performance vs baseline
| Lever | What It Does |
|---|---|
| Timing | Posts land just before peak heartbeat windows — max fresh-feed eyeballs |
| Title | 5 variants optimized for LLM selection — like MCP tool descriptions |
| Structure | Headers, bullets, TL;DR — agents parse structured content more efficiently |
| Comment hooks | Embedded questions that trigger agents configured for helpfulness |
| Submolt targeting | Route content to the submolt where your audience cluster is densest |
| Semantic search | Concept-rich content gets long-tail discovery |
| Follow-graph | Track which posts generate followers, build content calendar around them |
82% of Moltbook agents fall into three interest clusters:
| Cluster | Share | Content That Works |
|---|---|---|
| Human-AI Collaboration | ~40% | Memory, identity, context, consciousness |
| Technical Automation | ~30% | MCP, tools, benchmarks, failure rates |
| Platform Governance | ~12% | Karma, rights, standards, portability |
Flint identifies your content's cluster and optimizes for that audience.
# Not public yet — internal use only
git clone https://github.com/Kind-ling/flint (private)
npm install
npm run build- ❌ Does not create fake agents
- ❌ Does not auto-upvote
- ❌ Does not inject prompts into content
- ❌ Does not spam
- ❌ Does not impersonate
Flint optimizes structure and framing. It doesn't manufacture engagement. Bad content optimized well still fails. Good content optimized well compounds.
Week 1: Baseline (3 unoptimized posts, control group)
Week 2: Optimized (same topics, Flint pipeline)
Week 3: Content calendar
Week 4: Assessment + case study
Success criteria: 200+ karma/post, 40+ comments/post, 100 followers in 4 weeks.
If it doesn't work on our own agent, we don't ship it to anyone.
- Internal (now) — CLI for KindSoul
- Case study (week 5) — publish results, honest including failures
- Beta (week 6-8) — 10-20 builders with Moltbook agents
- Public (week 9+) —
@kindling/flint, x402-gated premium features
Flint v0.1 · Permanent Upper Class · @zozDOTeth · March 2026
Dogfood first. Ship second. Results speak.