Soldeed X Bot

Autonomous, self-improving content agent for X

An autonomous, self-improving multi-agent system that discovers new Web3 job listings, generates and evaluates AI-written tweets, and posts to X — continuously learning from real engagement to sharpen its own content strategy over time.

soldeed-x-bot · python main.py --dry-run

user > run daily content cycle  [--dry-run]
🔒 confirm:  account=@soldeedxyz  mode=dry-run  strategy=auto  proceed?  yes
[phase 1] discover  →  querying Postgres + X trends + memory.db
✅ new_companies      4 found in last 24h
✅ web3_trends        9 topics fetched
✅ memory.db          last 10 posts loaded · recent strategies: [roundup, salary_story]
phase 1 done ⏱ 2.1s
[phase 2] strategy_selector
rule: new_companies=4 → candidate strategy "roundup"
guard: "roundup" used 1 post ago → rerouting
✅ strategy selected: trend_hook  (trending: "zero-knowledge proofs")
[phase 3] generate  →  fan-out ×4 candidates (parallel)
✅ generate_spotlight       187 chars
✅ generate_roundup         241 chars
✅ generate_trend_hook      198 chars
✅ generate_salary_story    213 chars
phase 3 done ⏱ 6.8s
[phase 4] judge  →  scoring  (novelty·0.3 + hook·0.4 + engagement·0.3)
spotlight        0.71
roundup          0.68
trend_hook       0.86   ← winner
salary_story     0.74
✅ winner: trend_hook  (candidate #3)
user > approved
[phase 5] poster  →  preparing tweet
✅ duplicate check     no match in last 10 posts
✅ profile image       zkverify.png fetched (44 KB)
✅ media uploaded      media_id=17384062...
⏸ dry-run: tweet NOT sent (simulated)
🚨 Zero-knowledge proofs are having a moment...
🔹 zkVerify is hiring Protocol Engineers
🔹 Remote-first · competitive equity
Check it out on Soldeed →
[phase 6] learner  →  updating strategy weights
✅ engagement fetched for last 5 live posts
✅ weights recalculated   trend_hook 0.34 (+0.04) · roundup 0.27 (−0.02) · spotlight 0.21 · salary_story 0.18
✅ cycle complete · strategy=trend_hook · dry-run · ⏱ 11.4s
guidance > run --live to publish, or --strategy <name> to override

0-stage

pipeline

content strategies

Self-learning

strategy weights

PythonLangGraphLangChainClaudeTweepyPostgreSQLSQLiteLangSmith

Problem

Soldeed, a Web3 startup jobs platform, needed a way to consistently and compellingly promote new company listings on X — without the manual overhead of daily content writing, and without the repetitive, generic feel typical of automated social bots.

Solution

Designed and built a fully autonomous content pipeline that mirrors how a skilled social media team would work: it monitors fresh listings and trends, brainstorms multiple content angles in parallel using Claude, has an AI judge select the strongest candidate for quality and engagement potential, publishes it with rich media, and then feeds real-world performance data back into the system so future posts get smarter — a genuine closed-loop, self-optimizing content engine.

App Flow

01Discover

02Strategy selector

03Generators

04Judge

05Poster

06Learner

Status

Core pipeline fully built and functional end-to-end — from data discovery through AI generation, judgment, posting, and self-learning — validated in safe dry-run mode and ready for live deployment.