Acontext

Persistent memory and context infrastructure for AI agents. Remember, retrieve, and reuse context.

[Acontext @acontext_io]

$ACON WE'RE LIVE.

CA: 0x58ee94d32a0c1da11b2de28548fa3960acbc404e

Memory that evolves with every agent run.

Turning experience into reusable skills, making every next run smarter.

The learning loop starts here.

website : https://t.co/GLGZzt8CzP

[Acontext @acontext_io]

CASE #001: The agent that kept forgetting.

Evidence: 5 runs on the board. 2 failed.
Lead: the 3 that worked all point to one file, deployment-sop/SKILL.md.

Verdict: only the wins become skills. Case closed.

[Acontext @acontext_io]

Dev is back accumulating $ACON.

Another buy from the dev, showing continued commitment to the project and its long-term growth.

Building momentum and keeping $ACON moving forward.

.https://t.co/MGbIUDfyG0

[Acontext @acontext_io]

skill memory (n.)
/skɪl ˈmɛm.ə.ri/

The part of an AI agent that turns messy conversations into neat, reusable skill files, keeping only what worked.

See also: Acontext.

[Acontext @acontext_io]

What you see: your agent gets it right.

What's underneath: sessions and messages, task outcomes, your preferences, distilled skills, all kept as plain Markdown.

Acontext is the part below the waterline.

[Acontext @acontext_io]

My agent keeps forgetting ______.

Fill it in 👇

Acontext only keeps the pieces that fit: successful runs snap into a skill, the rest get tossed.

[Acontext @acontext_io]

An agent shouldn't run every lap alone.

Each run passes on what worked.
The next one starts ahead.
The baton gets brighter every handoff.

That's skill memory.

[Acontext @acontext_io]

Panel 1: agent runs npm install. Lockfile mismatch.
Panel 2: you: "we use pnpm here!"
Panel 3: it writes that down. Next session: pnpm install, first try.

That's the whole loop.

[Acontext @acontext_io]

Dev is back buying $ACON.

Continuing to support the project and strengthen its momentum.

The journey continues.

https://t.co/XJM6mvDvP9

[Acontext @acontext_io]

Every AI agent forgets what it learned.

Acontext captures every run, distills only what worked into plain Markdown skills, and hands them back so every next run starts smarter.

Agent skills as a memory layer. Open source.

[Acontext @acontext_io]

Think of your agent like a garden.

Every run is a seed. The ones that work grow into skills it can pick from later. The ones that fail don't take root.

Acontext does the gardening, so each season starts greener.

[Acontext @acontext_io]

A map of skills an agent can pick up:

🚀 deployment-sop
🧪 api-testing
🗄️ db-migration-sop
🔀 git-pr-workflows
📝 your preferences

Acontext writes islands like these from runs that worked. Save this for later.

[Acontext @acontext_io]

Self-hosting Acontext:

☑ curl -fsSL https://t.co/nMY814bfjC | sh
☑ database, storage, API and dashboard on your machine
☑ nothing leaves your servers

Your agents' context stays on your own servers. Open source.

[Acontext @acontext_io]

burned supply again.

https://t.co/6WS6a2Ltp1

[Acontext @acontext_io]

What would you want to see about your agent?

Acontext's dashboard tracks sessions, pulls tasks out of conversations, shows what succeeded and what failed, and lists the skills it learned.

Observability for what the agent did, not just the tokens it burned.

[Acontext @acontext_io]

Climb 1 → slips into 2 crevasses, still reaches the top, marks the trail.
Climb 2 → follows the markers straight up.

Acontext only learns from runs that succeeded, so the route it saves is the one that worked.

(illustration, not a benchmark)

[Acontext @acontext_io]

Hot take: an agent should get more useful every week you use it.

Most don't. Session 50 knows as much as session 1.

With skill memory, every successful task adds or improves a skill, so the library compounds.

[Acontext @acontext_io]

Where Acontext plugs in:

✓ Claude Code
✓ OpenClaw
✓ OpenAI SDK
✓ Anthropic SDK
✓ LangGraph
✓ Agno

Sessions in, skills out. One skill memory for the agents you already run.

[Acontext @acontext_io]

DEXPAID!

[Acontext @acontext_io]

Agent runs in.
Skills out.
Failed runs go in the bin. ✅❌

[Acontext @acontext_io]

Anatomy of a skill Acontext writes.

↳ use_when: the situation that should trigger it
↳ preferences: your rules, learned from your corrections
↳ tool_sops: which tools, in what order

Three fields. Plain Markdown. Your agent reads it before it starts.