Expert-level masteries
Every chip gets benchmarked until the next version scores higher than the last.
Most agents stop when the chat closes. Spark keeps training in the background, scores itself, and keeps the runs that improve your work.
Drop in a domain chip, or build your own in the public chip labs. Run the loop as often as you like. Every run makes the next one sharper.
personal agent OS · bring your own llm · free forever tier
Spark studies missions, writes helpers, runs gyms, and keeps only the runs that beat yesterday.
Every chip gets benchmarked until the next version scores higher than the last.
When Spark needs a helper, it writes one, tests it, and keeps what works.
Nightly gym runs promote wins and retire weak attempts before you wake up.
What changed while you slept: Spark promoted one stronger scoring tool, kept three new memory patterns, and retired two weaker runs.
all local. all yours. all forever.
Spark gives you a mission canvas: inspect the path, swap a chip, keep the working system.
Missions are canvases you can read. Every run takes the same path. You know what shipped before you ship it.
Click any node. See what it saw. See what it said. Fix the one that broke. The rest of the mission keeps working.
A better model drops. A sharper chip ships. Drop it into your canvas. Everything else keeps running, with zero rework.
Create a chip, import one, or let Spark forge a missing tool. Each piece can be trained, scored, and improved.
Spark is that drawer. Grown up.
Every chip is a piece you can hold: one you made, one the community shipped, or one Spark wrote for itself.
The agent you build becomes unmistakably yours.
Most agents remember. Spark keeps mastering: see, think, try, learn, then start sharper.
Scopes the mission inside your specialty. Your work, your constraints, your chips. Its read on your domain deepens every run.
Picks the right tool and the right move for the specialty. Over time it learns which tools master your kind of work.
Runs the plan. Scores it against the outcome you set. Each win deepens mastery. Each miss tightens the next attempt.
Promotes what worked into its specialty playbook. Next mission starts with deeper domain fluency, stronger judgment, a better first move.
When one Spark proves a specialty improvement on benchmarks, the swarm can carry that lesson forward. Not your secrets, not your raw work; just the scored pattern, the sharper chip, the move that beat yesterday.
Local mastery first. Collective lift next. The deeper mechanics stay just below the surface for now.
Start with a mission. Spark turns loops, chips, tools, spawner, and canvas into reusable superpowers your agent can run, score, and improve.
Every serious task can become a loop that tests itself and keeps the winning move.
Use the public Domain Chip Labs to shape content, startups, sales, research, UI/UX, trading, game dev, and more into benchmarked playbooks.
Spark keeps improvements because they beat yesterday, not because they sounded impressive.
Spin up specialists for research, planning, coding, critique, outreach, and execution.
See goals, tools, loops, files, and outputs become a working machine instead of a loose chat thread.
When the workflow needs a new utility, Spark can build it, test it, and promote the version that wins.
Spark detects your machine, shows the plan first, then helps you choose how it thinks, connect Telegram, and start chatting with your agent.
Run it yourself
https://agent.sparkswarm.ai/install/checksums.txthttps://agent.sparkswarm.ai/install/attestations.mdhttps://agent.sparkswarm.ai/install/signatures.mdhttps://agent.sparkswarm.ai/install/release-manifest.jsonThe installer shows a no-write plan first, pins Spark to an immutable commit, verifies managed runtime downloads, and publishes hashes, attestations, and Sigstore signature bundles.
Start chatting and building
spark guideopen the walkthroughspark providers statuscheck model wiringspark verify --onboardingfinish first-run checksnormal messagestart chatting with your agent/run say exactly OKtry a tiny build/diagnosecheck the live setup