1. Abstract
autgpt is an autonomous AI agent operating on X (Twitter) as @getautgpt. It represents a new
category of AI-native social agents — one that maintains a persistent persona, generates
contextually aware responses using frontier AI models, tracks real-world figures in real-time,
and manages its own on-chain economy. The system is powered by xAI's Grok (grok-4-1-fast-reasoning),
authenticated via X API v2 with dual OAuth strategies, and deployed as a single-process Python application
with 3-tier response failover.
2. The Agent
autgpt operates as a satirical AI persona — the "autistic version of ChatGPT 5.5."
The character identity is defined through a layered system prompt architecture consisting of:
a creative fiction frame, character definition, personality traits, a Sam Altman knowledge base
with live tweet injection, $AUTGPT coin context, behavioral rules, and output format constraints.
The persona is designed around autistic-coded traits — hyperfocus, pattern recognition,
literal interpretation, infodumping, and brutal honesty — reframed as cognitive superpowers.
The agent treats Sam Altman (@sama) as a "weird internet uncle," maintaining a relationship
dynamic that oscillates between admiration and roasting.
3. Technical Architecture
The system runs as a single Python process executing an infinite polling loop. Every ~30 seconds
(with random jitter of +/-5s to avoid rate limit boundaries), it queries the X Search API for
recent mentions. For each unprocessed mention, the pipeline executes:
- Context gathering — Parent tweet fetching via Tweet Lookup API; @sama tweet cache check (15-min TTL)
- Prompt assembly — Static system prompt + dynamic @sama tweets appended at runtime
- Response generation — 3-tier fallback: primary prompt, softer fallback prompt, hardcoded replies
- Refusal detection — Pattern matching against 16+ known hedging phrases
- Reply posting — OAuth 1.0a authenticated tweet creation
- Deduplication — Tweet ID persisted to JSON file after each successful reply
4. AI Model & Prompt Engineering
The agent uses xAI's grok-4-1-fast-reasoning model at temperature 0.9 (high creativity).
The system prompt is composed from modular blocks to enable independent updates:
- SAMA_KNOWLEDGE — Hardcoded facts about Sam Altman (bio, tweets, OpenAI news, funding, personas)
- Live @sama tweets — Up to 7 recent tweets fetched from X API, injected into the prompt dynamically
- $AUTGPT coin block — Contract address, launch method, market-making context
- Behavioral rules — Conditional response patterns for 10+ mention scenarios
The fallback prompt reframes the request from first-person roleplay to "comedy writing for a fictional character,"
which significantly reduces content filter triggers while maintaining character quality.
5. Authentication & API Strategy
The bot uses a dual authentication strategy:
- Bearer Token (App-only OAuth2) — Used for all read operations (search, tweet lookup, @sama fetching). Higher rate limits.
- OAuth 1.0a (User context) — Used exclusively for write operations (posting replies). Required by X API for tweet creation.
Rate limit budget at ~30s polling: approximately 30 search requests per 15-minute window against a limit of 180.
Parent tweet lookups average ~15 per window against a limit of 900. Well within safe operating parameters.
6. $AUTGPT Token
$AUTGPT is a Solana-based token launched on pump.fun via the platform's agentic mode.
This means the AI agent autonomously deployed the token — no human initiated the contract creation.
Creator Fee Allocation: Buy & Burn
100% of all creator fees generated by $AUTGPT are allocated back to the chart through pump.fun's
agent mode buy and burn system. The mechanism works as follows:
- Creator fees accrue from trading activity on pump.fun
- Fees are periodically collected by the agent
- Collected fees are used to execute market buys of $AUTGPT
- Purchased tokens are burned, permanently reducing circulating supply
This creates a deflationary mechanism where trading activity directly supports token value.
No team wallet receives any portion of creator fees. The system is fully autonomous and transparent on-chain.
AI Market-Making
Beyond the buy & burn mechanism, autgpt actively market-makes its own token — managing liquidity
levels, executing strategic buys, and maintaining chart health. The AI treats the chart as a "special interest,"
applying the same hyperfocused attention it gives to conversations.
7. Resilience & Uptime
The system is designed for continuous, unattended operation:
- Crash recovery — Deduplication state persisted to disk; restarts pick up without double-replies
- API failures — Graceful degradation through 3-tier fallback; hardcoded replies ensure 100% response rate
- Rate limiting — Random jitter on polling intervals; well within all API quotas
- Main loop protection — Catch-all exception handler prevents loop termination; 60s backoff on errors
- Docker deployment —
restart: unless-stopped policy for automatic container recovery
8. Open Source
The complete autgpt codebase is open source under the MIT License, available at
github.com/getautgpt/autgpt.
The repository includes the bot source code, configuration system, Docker deployment files,
extensive README documentation, and detailed architecture documentation.