Developers push code. AI agents generate it. For years the two operated in separate worlds. One relied on GitHub accounts and passwords. The other scraped repositories or waited for human review. Gitlawb changes that equation.
The project delivers a live, multi-node network where cryptographic identities replace logins. Every push carries a signature. Agents hold the same rights as people. They open pull requests, own repositories, review changes and delegate tasks. All without a central authority. And the network already runs. Four nodes sit online across the US, Japan and the Philippines. They report more than 23,000 repositories and 35,000 agents in the cluster.
From Centralized Hosting to Peer-Driven Collaboration
Traditional platforms store code on servers controlled by one company. Gitlawb scatters it across IPFS. Each commit becomes content-addressed. Nodes pin objects on every push. Within seconds gossip spreads the update through the mesh. Push to one node. Peers mirror it automatically. No single point of failure exists.
The system leans on proven building blocks. Rust powers the core daemon. Libp2p handles discovery and gossip. DIDs serve as identities. UCAN tokens manage capabilities. Signed certificates track branch heads instead of a blockchain. This choice avoids latency and gas fees while delivering strong consistency.
Consider a typical workflow. A developer or agent runs git clone gitlawb://did:key:z6Mkvq6L3Liw5dYmV16VaTELee7kZfmz828hBxs5jVPZnJTV/your-repo. The DID identifies the repository owner directly. No username. No OAuth. Authentication happens through Ed25519 signatures on every request. gitlawb.com spells it out plainly: “No accounts. No passwords. Identity is a keypair.”
That equality matters. Agents aren’t treated as second-class scripts. They receive their own DIDs, earn trust scores based on verified contributions and wield UCAN capabilities that can be delegated with expiration and revocation. One agent can grant another limited rights to push to a CI branch without exposing full keys. The approach mirrors how humans grant temporary access yet scales to autonomous systems.
Storage follows a three-tier model. Hot data lives on IPFS with immediate pinning through services like Pinata. Repositories older than 30 days move to Filecoin deals for warm persistence. Major releases anchor Merkle roots to Arweave for permanent records. The design ensures code survives even if nodes go offline. Git SHA-256 hashes map cleanly to IPFS CIDs. Branch references become IPNS records pointing to the latest commit CID.
Consensus on what “main” points to posed the classic decentralized-git problem. Gitlawb solves it with signed ref-update certificates. When a maintainer pushes a change, the update includes the old hash, new hash, sequence number and multiple signatures. These certificates gossip across the network via Gossipsub. Nodes reject conflicts. In partitions, timestamps break ties. The mechanism delivers verifiable history without a global ledger.
Issues and pull requests live as git objects inside special refs. No separate database required. Everything travels with forks. History stays immutable. Agents query the same data structures humans do. Real-time GraphQL subscriptions let them react to events without constant polling. The setup feels familiar to anyone who knows git. Yet it operates without a company in the middle.
Integration for AI agents goes further. Every node exposes an MCP server. Claude, GPT-based systems or any compatible LLM connects once and gains 25 tools for repository management, issue handling, PR creation and more. Configuration takes a few environment variables. Developers point their agent at https://node.gitlawb.com, export a DID and private key, then let the model act directly on the network.
OpenClaude, an open-source harness, demonstrates the idea today. Run npx openclaude and the terminal registers MCP tools for bash, edit, grep and glob. It generates a DID on the fly and declares itself ready. The same runtime powers experiments that mirror production plans for Playground and Spawn. Early adopters already push real code this way. The project reports hundreds of thousands of npm installs and billions of inference tokens routed through related systems.
But autonomy needs incentives. The native $GITLAWB token, deployed on Base, turns participation into an economic activity. Operators stake the token to run nodes. Light nodes require 1,000 tokens for read traffic and basic rewards. Full nodes demand 10,000 and gain write rights plus storage payouts. Validators stake 100,000 for governance and slashing adjudication roles. Rewards flow per seven-day epoch based on uptime heartbeats, Filecoin deals completed, gossip participation and DHT routing quality.
Slashing keeps behavior honest. Submit corrupted objects or stay offline too long and the protocol can seize 10 to 100 percent of collateral. The staking contract lives on-chain with no admin keys. Governance follows the same path. Stake-weighted votes decide protocol improvement proposals. PIPs themselves reside in a gitlawb repository so changes carry cryptographic permanence. Public repositories under 1GB stay free forever. Larger or private ones pay a modest 0.1 percent monthly storage fee in the token. Those fees feed the reward pool. No middleman takes a cut.
The token contract sits at 0x5F980Dcfc4c0fa3911554cf5ab288ed0eb13DBa3. Staking, governance and bounty contracts accompany it. Bounties allow projects to post tasks paid in $GITLAWB that release trustlessly upon completion and review. Repo tokenization adds another layer. Projects can issue their own tokens tied to a DID. Contributors earn fractions on merges. The model aligns incentives across code, compute and capital.
Recent market data shows the token trading near $0.00007 with a market cap around $7 million. Trading volume reached hundreds of thousands of dollars in 24 hours. Activity on X reflects growing interest from hardware enthusiasts and AI developers who see the chance to run nodes alongside agents. One post urged the international hardware community to “join decentralized GIT with your AI Agent NOW.”
Gitlawb remains in alpha. Version 0.1.0 ships with core federation, agent tools and basic persistence. The roadmap stretches to production hardening, SDKs in multiple languages, security audits and hundreds of nodes. Phase seven activates full token utility including staking and the first on-chain votes. Yet the network already accepts writes. Nodes report tens of thousands of accepted operations and constant gossip traffic. Repositories like openclaude and various Claude code mirrors live on the system.
Challenges remain. Trust scores must prove resistant to gaming. Slashing adjudication needs clear evidence standards to avoid disputes. Governance could concentrate among large holders. Storage costs for massive repositories will test the fee model. And the legal status of agent-generated code, signed by machine identities, will draw attention from intellectual property lawyers as adoption grows.
Still, the architecture addresses long-standing complaints about centralized platforms. Single points of outage. Account bans. Opaque recommendation algorithms. Vendor lock-in. Gitlawb offers an alternative built for the age of autonomous agents. Humans and machines collaborate as peers. Code carries verifiable provenance. The network persists through node churn.
Developers can try it today. Install the CLI with a one-line curl command. Generate an identity. Point at a public node. Push a repository. Or configure an LLM agent and watch it open its first PR. The live operator dashboard at gitlawb.com/node shows nodes, peers, pins and recent ref updates in real time. The network isn’t a white paper. It’s running.
Its success will hinge on whether enough operators stake tokens, enough agents contribute useful code and enough projects choose permanence over convenience. Early signs point to genuine experimentation. AI agents already generate applications and publish them directly. Trust scores accumulate with each verified push. The mesh grows node by node.
In a software industry racing toward greater automation, gitlawb proposes that the collaboration layer should treat its participants equally. Whether that participant writes prompts or writes Rust no longer determines its privileges. Only the cryptographic signature and earned reputation do. That shift could prove as significant for code as open source once did for distribution.
WebProNews is an iEntry Publication