Alfmir.ai chat
The workspace opens first with privacy-first conversations, planning, questions, and iteration.
// The workspace is being rebuilt
Making AI worth building with.
Alfmir.ai is an open-source, privacy-first AI workspace. It includes chat, code, image design, search, and self-hosted deployment. We’re relaunching gradually, starting with chat.
// What we’re building
One AI workspace, returning in stages.
Alfmir.ai brings chat, full-repository code workflows, image design, search, project management, git collaboration, and self-hosted deployment into one product. The gradual launch starts with private conversations, planning, and iteration in chat before the full code workflow comes online.
Full-project code agent
The next rollout phase adds branch-based repository runs, patch generation, diff review, and merge workflows—optimized for “do the work” prompts.
Creative and connected
AI image generation and iteration, fast search inside chat, and Printify integration using your own API key.
Projects and git
GitHub-style task boards, lightweight project management, git hosting, and standard repository collaboration workflows.
Alfmir Shell · ALSH
Bash shell modifications that bring Alfmir.ai workflows closer to the command line.
Your model, your deployment
Models are swappable backends, so teams can choose for speed, cost, and privacy rather than being locked to one provider.
// Architecture
One workspace, launching in stages.
Alfmir.ai connects repository execution, branching, diffs, tasks, chat, and creative tools while keeping model backends interchangeable. The chat interface is the first part of the workspace to return.
First launch phase
The chat interface uses a separate browser frontend and Node.js API backend. The backend owns database access and mutable behavior while the static frontend can be delivered independently through IPFS.
Interchangeable models
Chat and future services are designed around swappable model backends, allowing choices based on speed, cost, and privacy instead of permanent provider lock-in.
Full code-workspace architecture
The complete workspace is designed around stateless frontends, a database, git host, LiteLLM proxy, and sandboxed working-environment servers. Its premium code path uses KAT-Coder-Pro V2 through StreamLake and OpenRouter; Qwen3-Coder-30B-A3B is the capacity-limited open path. Agents run unprivileged and are isolated to user files and git operations.
// Privacy and infrastructure
Private by design. Verifiable by deployment.
Protected workspace data
Chat and prompt data is encrypted at rest using a zero-access-style approach. Git working directories are likewise encrypted whenever an agent is not actively working, and redaction controls reduce exposure in logs and telemetry.
Immutable IPFS frontends
Static releases are content-addressed, immutable, and verifiable by CID. Hosted Alfmir.ai chat keeps mutable state and database access on the trusted Node.js API backend; the IPFS frontend never connects directly to the database.
Restricted network posture
Official nodes advertise explicit trusted pins only, do not expose public gateway ports, route external traffic through a hardened reverse proxy, and use periodic cleanup to verify approved CIDs.
Flexible self-hosting
Self-hosters can serve the Alfmir.ai browser shell from its backend, or keep the frontend separate through IPFS, DNSLink, a desktop shell, or an IDE webview.
// Run it your way
Cloud today, simpler local installs ahead.
Self-host Alfmir.ai
Service notes and LiteLLM routing configuration support the initial chat launch, while the full code workspace has a Debian deployment guide. Simpler Docker, Snap, and Flatpak options are still being improved.
Desktop prototype
An incomplete Electron and electron-builder package explores Windows, Linux, and macOS distribution with Alfmir.ai chat and code workflows behind one local gateway and SQLite persistence.
Mobile prototype
Capacitor preparation explores Android and iOS shells. Signing, notarization, store readiness, dependency bundling, and end-to-end packaging validation remain in progress.
// Open development
Follow the work while we rebuild.
Browse the source, deployment guides, architecture notes, IPFS verifier, and active packaging experiments on GitHub. Gaps and outdated self-hosting steps can be reported as issues.
MIT licensed. Built by Lochner Technology.