惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

V
Visual Studio Blog
Engineering at Meta
Engineering at Meta
月光博客
月光博客
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
T
Tailwind CSS Blog
博客园 - Franky
The GitHub Blog
The GitHub Blog
大猫的无限游戏
大猫的无限游戏
The Cloudflare Blog
B
Blog RSS Feed
云风的 BLOG
云风的 BLOG
小众软件
小众软件
罗磊的独立博客
Microsoft Azure Blog
Microsoft Azure Blog
I
InfoQ
美团技术团队
H
Hackread – Cybersecurity News, Data Breaches, AI and More
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
V
V2EX
C
Check Point Blog
WordPress大学
WordPress大学
博客园 - 【当耐特】
博客园 - 司徒正美
D
Docker

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - sqliteai/adam: An embeddable cross-platform AI a...
marcobambini · 2026-05-06 · via Hacker News - Newest: "AI"

Embeddable AI agent library in C.

Adam gives you a complete agent loop: tool calling, memory, sessions, voice, streaming, structured output, in one #include. Works with cloud APIs (Anthropic, OpenAI, Google Gemini, Groq, Together, xAI) and local models (llama.cpp) through the same interface. Compiles on macOS, Linux, Windows, iOS, Android, and WASM.

Quick Start

#include "adam.h"

int main(void) {
    adam_init();

    adam_settings_t *s = adam_create_settings();
    adam_settings_set_provider(s, ADAM_API_ANTHROPIC,
                               getenv("ANTHROPIC_API_KEY"),
                               "claude-sonnet-4-20250514");

    adam_history_t *h = adam_history_create();
    adam_run_result_t r = adam_run(s, h, "What is the capital of France?");

    printf("%s\n", r.final_response);  // "The capital of France is Paris."

    adam_run_result_free(&r);
    adam_history_destroy(h);
    adam_settings_destroy(s);
    adam_cleanup();
}
make deps    # build llama.cpp + whisper.cpp
make all     # build libadam.a
make test    # run 161 tests (ASan + UBSan)

Features

Feature Description
Agent loop Tool calling with automatic iteration until final response
Three providers Anthropic, OpenAI, Google Gemini + any compatible API + local GGUF via llama.cpp
Local vision Multimodal image understanding via llama.cpp + mmproj (Gemma 3, LLaVA, etc.)
Image generation Native image output via Gemini image models (gemini-3.1-flash-image-preview)
Database extensions SQLite and PostgreSQL extensions — embed Adam as SQL functions that query the same database
13 built-in tools File I/O, shell, calculator, SQL, web fetch/search, HTTP POST, memory, research, multi-agent
Long-term memory Hybrid BM25 + vector search via SQLite (sqlite-memory + sqlite-vector)
Session persistence Save/load conversations with UUIDv7 keys
Telegram bot Full-featured Telegram integration with text, voice, images, tools, and memory
Voice STT (Whisper cloud/local) + TTS (cloud/system) + full audio pipeline
Streaming Real-time token delivery via callback
Structured output adam_run_json() with validation and retry
Evolution loop Self-improving agent: iterate, score, refine strategy
Research mode Autonomous multi-iteration information gathering with report synthesis
Multi-agent Agent A invokes Agent B as a tool, with independent settings/tools
Guardrails Pre-send and post-receive validation callbacks
Response cache LRU hash table keyed on model + message history
History management Clone, summarize (LLM-based compression), token estimation
Thread pool Concurrent agent execution with job queue
Filesystem sandbox Tools restricted to explicitly allowed directories
Cross-platform macOS, Linux, Windows, iOS, Android, WASM (Emscripten)
Arena allocator Zero-leak per-iteration memory with automatic cleanup

Build

make deps          # Build llama.cpp, whisper.cpp (+ mbedtls/curl on Linux)
make all           # Build libadam.a
make test          # Build & run unit tests (ASan + UBSan)
make chat          # Interactive text chat (cloud API)
make chat GGUF=models/model.gguf  # Interactive text chat (local)
make vision GGUF=models/model.gguf MMPROJ=models/mmproj.gguf  # Local vision test
make talk          # Voice agent (cloud)
make talk LOCAL=1  # Voice agent (fully local)
make memory        # Memory system tests
make clean         # Remove all build artifacts

API Reference

Full API documentation with every function, type, and callback: API.md

Examples

adam_settings_t *s = adam_create_settings();
adam_settings_set_provider(s, ADAM_API_ANTHROPIC, api_key, "claude-sonnet-4-20250514");

adam_history_t *h = adam_history_create();
adam_run_result_t r = adam_run(s, h, "Explain quantum entanglement simply.");
printf("%s\n", r.final_response);
adam_run_result_free(&r);

// Continue the conversation — history carries forward
r = adam_run(s, h, "Can you give an analogy?");
printf("%s\n", r.final_response);
adam_run_result_free(&r);

adam_history_destroy(h);
adam_settings_destroy(s);

More examples are available in the examples/ directory:

Example Description
simple-conversation Multi-turn chat with a cloud API
tool-calling Register a custom tool and let the agent call it
local-model Run a local GGUF model via llama.cpp
local-vision Image understanding with a local vision model + mmproj
google-gemini Use Google Gemini models
image-generation Generate images with Gemini
sqlite-query Natural language queries on any SQLite database
structured-json Get validated JSON output with retry
memory Long-term memory with hybrid BM25 + vector search
sessions Save and restore conversations
streaming Real-time token streaming via callback
voice Speech-to-text + agent + text-to-speech pipeline
multi-agent Agent A delegates to Agent B as a tool
evolution Self-improving agent with scoring loop
research Autonomous multi-iteration research with report
filesystem-sandbox File and shell tools restricted to allowed directories
guardrails Pre-send and post-receive validation
response-cache LRU cache for repeated queries
thread-pool Concurrent agent execution
telegram Telegram bot with text, images, tools, and memory
wasm-chat Browser-based chat UI via WebAssembly
full-agent All features combined

Database Extensions

Adam can be embedded directly inside SQLite and PostgreSQL as a SQL extension. The agent can query the same database it's loaded in — ask questions in natural language, get answers from your data.

-- SQLite
.load adam

-- PostgreSQL
CREATE EXTENSION adam;

-- Configure (both)
SELECT adam_config('provider', 'anthropic');
SELECT adam_config('api_key', 'sk-ant-...');

-- Ask about your data — the agent reads the schema and runs SQL
SELECT adam_ask('How many users signed up last month?');
-- → "47 users signed up last month."

-- Generate SQL without executing
SELECT adam_sql('top 5 products by revenue');
-- → "SELECT p.name, SUM(oi.quantity * oi.price) AS revenue FROM ..."
Function Description
adam_config(key, val) Configure provider, API key, model (persisted)
adam(msg) Stateless one-shot chat
adam_ask(msg) SQL-aware agent — reads schema, queries data, multi-turn
adam_sql(question) Generate SQL from natural language
adam_create_session() Create session (auto-created on first adam_ask)
adam_get_session() Get current session UUID
adam_clear_session() Clear session and history

See extensions/sqlite/ and extensions/postgres/ for build instructions.

Architecture

adam_run() loop:
  build system prompt (identity + instructions + bootstrap files + memory + datetime)
  -> check guardrails (on_before_send)
  -> check cache
  -> dispatch LLM (mock | local/llama.cpp | remote/HTTP)
  -> check guardrails (on_after_receive)
  -> if tool_calls: execute tools -> append results -> loop
  -> if text: return final response -> auto-save session -> extract memory

Platform abstraction: macOS uses NSURLSession, Linux uses libcurl+mbedtls, WASM uses embedder-provided http_fn callback.

Memory management: Arena allocators for per-iteration zero-copy work. malloc/free for long-lived structures. Arena-allocated strings are only valid within the current iteration.

Feature Gates

Define before #include "adam.h" to disable features:

Gate Effect
ADAM_NO_CURL No libcurl (must provide http_fn callback)
ADAM_NO_LOCAL No llama.cpp (no local inference)
ADAM_NO_PTHREADS No thread pool, no voice thread
ADAM_NO_SQLITE No SQLite (no memory, sessions, or SQL tool)
ADAM_NO_VOICE No voice subsystem
ADAM_NO_FILESYSTEM No file_read/file_write/list_directory tools
ADAM_NO_SHELL No shell_exec tool

Dependencies

All vendored as git submodules in modules/:

Module Purpose
llama.cpp Local LLM inference + GGML compute
whisper.cpp Local speech-to-text (shares ggml via symlink)
miniaudio Cross-platform audio I/O
sqlite Amalgamation build
sqlite-memory Hybrid BM25 + vector knowledge store
sqlite-vector Vector similarity search
mbedtls TLS (Linux only)
curl HTTP (Linux only)

macOS uses system frameworks (Foundation, Security, Metal, AVFoundation, Accelerate) instead of curl/mbedtls.

License

MIT