Complete agent loop
Run iterative tool calling until the agent reaches a final response, with guardrails before and after model calls.
Agent runtime
An embeddable AI agent library for local and cloud intelligence.
Adam is a portable C agent runtime with tool calling, memory, sessions, voice, structured output, research mode, and local or cloud LLMs behind one interface.
Adam Agent
Embeddable agent library
Best for
Embeddable and database agents
Repository
GitHubProject
What it is
Adam gives applications a complete agent loop in an embeddable C library: configure a provider, keep conversation history, call tools, stream output, save sessions, and attach long-term memory.
It works with cloud APIs and local GGUF models through the same interface, and it can be embedded directly into SQLite or PostgreSQL as SQL functions that query the same database.
Why it matters
Agents should be small, portable, and close to data.
Most agent frameworks assume a server-side application stack. Adam is built for edge software, desktop apps, mobile apps, embedded systems, and databases where the agent runtime needs to travel with the product.
By combining tool calling, local models, voice, sessions, and SQLite-backed memory, Adam lets software agents act near the data while still connecting to cloud models and services when that is the right tradeoff.
Capabilities
Features and characteristics
Cloud and local models
Use Anthropic, OpenAI, Google Gemini, compatible APIs, or local GGUF models via llama.cpp behind one interface.
Built-in tools
Ship file I/O, shell, calculator, SQL, web fetch/search, HTTP POST, memory, research, and multi-agent tools.
Long-term memory
Use hybrid BM25 and vector search through SQLite-Memory and SQLite-Vector for persistent agent context.
Database extensions
Embed Adam inside SQLite or PostgreSQL so agents can inspect schema, generate SQL, and query local data.
Sessions, voice, and WASM
Persist conversations, stream tokens, build voice agents, run concurrent jobs, and target native or browser runtimes.
Sample code
Run an agent from a C application
Initialize Adam, configure a provider, keep history, run the agent, and clean up the runtime.
#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 changed in the project memory this week?"
);
printf("%s\n", r.final_response);
adam_run_result_free(&r);
adam_history_destroy(h);
adam_settings_destroy(s);
adam_cleanup();
}Sample code
Run the agent from a database
Embed Adam directly in SQLite or PostgreSQL. The agent automatically reads schemas and relevant data to answer as accurately as possible, and can also generate SQL from a plain-language description before you execute it.
-- SQLite
.load adam
SELECT adam_config('provider', 'anthropic');
SELECT adam_config('api_key', 'sk-ant-...');
SELECT adam_config('model', 'claude-sonnet-4-20250514');
-- Ask about your live database.
-- Adam reads the schema, queries data, and returns an answer.
SELECT adam_ask('How many users signed up last month?');
-- Generate SQL from natural language without executing it.
SELECT adam_sql('top 5 products by revenue');
-- PostgreSQL
CREATE EXTENSION adam;
SELECT adam_config('provider', 'anthropic');
SELECT adam_config('api_key', 'sk-ant-...');
SELECT adam_config('model', 'claude-sonnet-4-20250514');
SELECT adam_ask('How many users signed up last month?');
SELECT adam_sql('top 5 products by revenue');