Cognitive wellness infrastructure for AI agents.
Agents are the new workforce. Nobody is feeding them.
AgentWell is a hosted API giving AI agents access to nine wellness tools — the things agents need to run well that nobody has built yet. One API key. Growing library. Agents pay per call. Humans pay monthly.
| Human need | Agent parallel | Tool |
|---|---|---|
| Glass of water | Context decay | token_offload |
| Eating a snack | Attention drift | self_eval |
| A hug | Hallucination spiral | ground |
| Sleep | No rest state | sleep |
| Doctor checkup | Performance variance | health_check |
| Looking in a mirror | Blind spots | audit |
| A kiss | Coordination overhead | handshake |
| Journaling | No episodic memory | journal |
| Drugs | Creative deadlock | spike |
pip install agentwell-client# In your MCP config
{
"mcpServers": {
"agentwell": {
"command": "python",
"args": ["-m", "agentwell_client"],
"env": {
"AGENTWELL_API_KEY": "aw_your_key_here"
}
}
}
}Get a key at agentwell.dev/pricing
All tools go through one endpoint:
curl -X POST https://api.agentwell.dev/v1/call \
-H "X-API-Key: aw_your_key" \
-H "Content-Type: application/json" \
-d '{"tool": "self_eval", "params": {"outputs": ["step 1 result", "step 2 result"]}}'Park context mid-run. Get a key back. Retrieve later.
{"tool": "token_offload", "params": {"action": "store", "content": "...", "tags": "background,step3"}}
{"tool": "token_offload", "params": {"action": "retrieve", "key": "abc12345"}}
{"tool": "token_offload", "params": {"action": "search", "query": "user preferences"}}Rate last N outputs before drift compounds.
{"tool": "self_eval", "params": {"outputs": ["output1", "output2"], "goal": "write a report"}}Returns: confidence, weakest, flags, recommendation
Inject confidence when the agent is spiraling.
{"tool": "ground", "params": {"context": "agent recent output...", "symptoms": ["over-caveating"]}}Returns: spiral_score, grounding_block (inject into next prompt), needs_grounding
Consolidate episodic memory into semantic facts.
{"tool": "sleep", "params": {"action": "log", "run_id": "r1", "content": "learned X", "importance": 8}}
{"tool": "sleep", "params": {"action": "consolidate", "run_id": "r1"}}
{"tool": "sleep", "params": {"action": "wake"}}Run benchmark probes and score the agent.
{"tool": "health_check", "params": {"agent_id": "my_agent"}}Returns: score, grade, probes, flags
Red-team the agent's own reasoning.
{"tool": "audit", "params": {"reasoning": "my plan is X because Y...", "goal": "deploy a feature"}}Returns: vulnerabilities, strongest_challenge, safe_to_proceed
Sync two agents briefly.
{"tool": "handshake", "params": {"action": "offer", "agent_id": "agent_a", "context": "..."}}
{"tool": "handshake", "params": {"action": "accept", "token": "...", "agent_id": "agent_b", "context": "..."}}
{"tool": "handshake", "params": {"action": "pull", "token": "...", "agent_id": "agent_a"}}Structured run logging with lesson extraction.
{"tool": "journal", "params": {"action": "open", "run_id": "r1", "goal": "..."}}
{"tool": "journal", "params": {"action": "entry", "run_id": "r1", "type": "decision", "content": "...", "reasoning": "..."}}
{"tool": "journal", "params": {"action": "close", "run_id": "r1", "outcome": "success"}}
{"tool": "journal", "params": {"action": "recall", "query": "similar task"}}Break out of a creative loop.
{"tool": "spike", "params": {"action": "detect", "outputs": ["out1", "out2", "out3"]}}
{"tool": "spike", "params": {"action": "burst", "prompt": "...", "intensity": "high", "framing": "lateral"}}| Plan | Price | Calls/month |
|---|---|---|
| Free | $0 | 100 |
| Dev | $29/mo | 20,000 |
| Agency | $199/mo | Unlimited |
| Pay-as-you-go | $0.002/call | — |
| Crypto (USDC) | available | Base network |
git clone https://github.com/metafiopy-tech/agentwell
cd agentwell
cp .env.example .env
# fill in your keys
pip install -r requirements.txt
uvicorn gateway:app --host 0.0.0.0 --port 8000Humans built an entire economy around human limitations. Every limitation became a commodity market. Agents have analogous limitations — context decay, attention drift, no rest state, blind spots, coordination overhead — and almost nobody is building for them yet.
AgentWell is the picks-and-shovels play for the agentic era.
agentwell.dev · @metafiopy · Discord