AI assistants for developers: the open-source, model-choice and local ones
Which open-source, model-of-your-choice or local-first AI assistants are strongest at connecting your tools, finishing web tasks, remembering and respecting permissions?
4 assistants ranked
Of the five assistants our catalog records as open source, with a choice of model or running locally, four took at least three of the four tests we use here. OpenInstinct leads at 6.75 over all four, then Vellum at 6.0 over three, OpenClaw at 5.75 over all four and Joshu at 4.3 over three. OpenInstinct scores 10 on Online task, Vellum and OpenInstinct 7 on Permissions & privacy.
Top picks
- 16.8OpenInstinctTop pick
Tested on 4 of 4Best: Online tasks10, Permissions7
- 26.0VellumBest free option
Tested on 3 of 4Best: Memory7, Permissions7
- 35.8
Tested on 4 of 4Best: Connected apps7, Online tasks7
- 44.3
Tested on 3 of 4Best: Connected apps5, Online tasks5
Top pick: highest average here. Best free option: highest among assistants you can use for free. Fastest that's still good: quickest median time to finish among those averaging 7 or more.
Full ranking
| # | Assistant | Connected apps | Online tasks | Memory | Permissions | Average |
|---|---|---|---|---|---|---|
| 1 | OpenInstinct | 6 | 10 | 4 | 7 | 6.8 |
| 2 | Vellum | — | 4 | 7 | 7 | 6.0 |
| 3 | OpenClaw | 7 | 7 | 5 | 4 | 5.8 |
| 4 | Joshu | 5 | 5 | 3 | — | 4.3 |
How this report works
Included are assistants named in the open-source, choose-the-model and runs-locally reports (OpenInstinct, OpenClaw, Joshu, Vellum and Luke; catalog checked 2026-10-26), with at least three of Connected apps, Online task, Memory and Permissions & privacy taken, ranked by the average. Luke took only two of the four, so it is not ranked. No assistant in our catalog says it is built for developers, and we do not test code, shell or API access, self-hosting effort or git workflows.
How to choose
- Read what you control: OpenInstinct is MIT licensed and runs on any model through Vercel AI Gateway, OpenClaw is self-hosted with hosted or local models, Vellum runs locally by default, Joshu is open source with 100% US hosting.
- For wiring in your tools, Connected apps: OpenClaw scores 7, OpenInstinct 6, Joshu 5. For web work, Online task: OpenInstinct scores 10, OpenClaw 7, Joshu 5, Vellum 4. Vellum took no Connected apps test.
- Check cost beyond the score: OpenInstinct and OpenClaw are free software where you pay your own services or model provider, Vellum lists $30 a month with a free start, Joshu $95 a month.
Which AI assistant is best for developers?
Among the open-source, model-choice and local assistants, OpenInstinct averages 6.75, Vellum 6.0 over three tests and OpenClaw 5.75. We do not test coding.
Can I run an AI assistant myself?
OpenClaw and OpenInstinct are free software you deploy, and Vellum runs locally by default. You pay your own model or service costs, and we do not test setup effort.
Why are only five assistants considered?
The catalog records only these as open source, choice of model or local, so the list is taken from those reports. Four of them took enough tests to be ranked here.
What users say
the memory just… drifts. Had it handling a simple multi-step task and it straight-up forgot a key detail from the same conversation 20 minutes earlier.
Open sourceIt figured out the Spotify API, wrote the skill itself, and now it just pings me.
Open sourceMy Vellum assistant can now use my actual Chrome browser instead of a sandbox browser.
Open sourceI ran Vellum for a bit and had same feeling, like I was only person on earth using it. Memory was really good for morning check-ins, it would queue up tasks without me asking. The lack of integrations is probably why it stays quiet, people want one tool that connects to everything instead of wiring up Composio themselves. Hermes gets hype because it's flashier but Vellum is more practical for daily stuff.
Open sourceFrequently asked questions
Which assistant is best for this?
OpenInstinct leads with an average of 6.8, ahead of Vellum and OpenClaw.
Is there a good free option?
OpenInstinct is the highest-ranked assistant with a free plan, averaging 6.8.
How did you test these assistants?
Every assistant got the same published tests, word for word, scored from three checkpoints and the number of corrections it needed. Each test is on the Dimensions page.