The problem with AI agents isn't that they forget — it's that they improvise. durin is built the other way around: it remembers your work, sharpens its own skills, and runs on adaptable workflows that make it follow the exact steps you define instead of inventing its own. Reliable execution you control — local-first, on any model.
durin is named for Tolkien's dwarf-king of Khazad-dûm; the mark is a dwarven anvil lit by a forge flame.
Why durin
Workflows that follow your steps
Left to itself, an LLM agent improvises a different path every time — great for open-ended chat, unreliable for work you need done a specific way. durin lets you pin it to adaptable workflows: visual graphs of nodes with routing, loop-backs, parallel branches, sub-flows, and human checkpoints. A task then runs the way you designed it — every time. The agent can pause to ask you, resume exactly where it stopped, and even edit and improve a workflow on its own (with an auto-revert safety net). Few personal agents give you this much control over how the work gets done.
Automations that fire on their own
A workflow runs once when you ask; an automation runs it again on its own — on a schedule, an inbound message, a webhook, or another automation finishing. durin classifies the result, decides whether it's worth telling you about, and chains into the next automation when one is configured. When a run needs a decision only you can make — an approval, a question — it pauses and asks, either you or the counterpart you're corresponding with, instead of guessing. An automation can even carry a goal — "until invoice 4471 is paid" — and switch itself off the day it's reached.
A memory that's actually yours
durin builds an entity graph from your conversations — people, projects, documents, decisions — and keeps it as plain markdown you own. Recall is instant and offline: it searches an FTS + vector index without an LLM in the hot path, so it's fast and private. Documents you hand it (PDF, Office, EPUB, web pages) land in a separate Library, kept apart from everyday recall. The markdown is the source of truth; the search indexes are derived and rebuildable. (Benchmarked at LoCoMo ≈ 0.79.)
A skills ecosystem with guardrails
durin discovers skills from public marketplaces, security-scans each one for scam/malware patterns before you run it, and connects MCP servers from a curated catalog with one-click install and auto-detected OAuth. Its composition doctrine keeps behaviour reliable — prefer deterministic code, use workflows for structure, treat skills as the through-line the agent follows — and it sharpens its own skills over time.
See exactly what it's running
Every run — workflows, scheduled cron jobs, agent tasks — shows up in a live panel with per-step detail, status, and the model used. Nothing happens in a black box: watch a run, open a paused checkpoint, and pick up right where it stopped.
And more
- Any model, or none. 35 provider integrations — Anthropic, OpenAI, Gemini, NVIDIA, Mistral, DeepSeek, Groq, Z.AI, Qwen, Moonshot, and more, plus OpenRouter and other gateways that reach hundreds of models. Switch mid-conversation, or run fully local with Ollama / llama.cpp / vLLM / LM Studio.
- Talk to it anywhere. A CLI, a Textual TUI, a browser dashboard, and chat channels — Telegram, Slack, Discord, email, WhatsApp, WebSocket — all reach the same agent and the same memory.
- Local-first & private. Runs as a daemon on your own machine under
~/.durin/; nothing phones home. - Personas, voice, and schedules. Per-channel personas (a SOUL + a model), local speech-to-text and text-to-speech, and cron for scheduled work.
…and it gets sharper while you sleep
A cold-path dream runs offline: it consolidates your conversations into the memory graph and curates your skills, so durin quietly gets better at helping you over time — without you lifting a finger.
Quick start
durin installs with uv, which fetches a compatible
Python for you — nothing on your system has to match. Don't have uv yet?
curl -LsSf https://astral.sh/uv/install.sh | sh
uv tool install --prerelease allow durin-agent # PyPI (uv picks a compatible Python)
durin onboard # interactive setup wizard
durin doctor # confirm setup is healthy
durin agent # launch the TUI
Run the gateway for the browser dashboard plus chat channels:
durin gateway start # dashboard at http://127.0.0.1:8765
See the install guide for prerequisites, optional extras (memory, local models, audio), and platform notes. For everyday and in-session commands, see the CLI reference.
Documentation
- Install · configure · uninstall
- CLI & in-session commands
- Configuration reference
- Providers & models
- Channels — Telegram, Slack, Discord, email, and more
- Documents & your knowledge
- Workflows
- How it works (internals)
License
MIT
1 maintainer
Release files for durin-agent 0.9.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| durin_agent-0.9.3.tar.gz | 33.2 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| durin_agent-0.9.3-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 66.6 MB
Release files / durin_agent-0.9.3.tar.gz
| Download URL | durin_agent-0.9.3.tar.gz |
|---|---|
| Size | 33.2 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
c9186c6c35cdd7fadac26afebee45ef5d85d6d8a17cf0b90f3cf434fc854a31e
|
|
BLAKE2b-256 checksum How to use checksums |
9b921311ccecb697a5b945fbda5f385ea2059e2e36c734e56af84dca2285e933
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 7, 2026.
Transparency logRelease files / durin_agent-0.9.3-py3-none-any.whl
| Download URL | durin_agent-0.9.3-py3-none-any.whl |
|---|---|
| Size | 33.4 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
01fb07c0dc599656df5c2feaa26c5b0cfd9b1ccce348dbb38815f8aac9b515d5
|
|
BLAKE2b-256 checksum How to use checksums |
6f20d72d2d8ab191053bc076d352d01bf479d687e4987d1b06675411756175ab
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 7, 2026.
Transparency log
