Rating certificate · AW/26-09/02226 Sept 2026
Not enough evidence
The Register / Orchestrators & meta-harnesses / Multica
Register entry · Orchestrators & meta-harnesses · updated 26 Sept 2026
Multica
Multica is a source-available workspace in which agents such as Claude Code, Codex, Cursor and Copilot appear as assignees on an issue board. A daemon on the user's machine or server picks up assigned issues, runs the agent CLI in a per-run working directory, posts progress comments and returns a pull request for human review. It adds squads, reusable skills, cron-style autopilots, execution logs and token accounting, and can be self-hosted with Docker Compose or Helm or used via the hosted multica.ai service.
What reviewers and users praise
- Reviewers see it filling a real coordination gap for teams running several agents (AgentConn, ArshTechPro)
- Users praise the kanban/assignee model as easy to pick up (X, Reddit)
- Self-hostable with agents running on the user's own machines (Flowtivity, docs)
What they criticise
- Early software with fast-moving versions and upgrade regressions (GitHub discussions, ArshTechPro)
- Users report high token consumption and heavy per-run disk use (GitHub #3292, discussion #6709)
- Agents run with the daemon user's full permissions and Claude Code is launched in bypassPermissions mode; isolation is left to the operator (security-model docs, source)
- Enterprise features such as SSO and audit logs were still on the roadmap per a spring 2026 review (Flowtivity)
Experts vs the crowdneeds both scores
Experts—
Crowd—
| Price | Free |
|---|---|
| Use it via | Web app, Desktop app, CLI, API, Chat apps |
| Runs on | macos, windows, linux, web |
| Licence | LicenseRef-Multica (Apache-2.0 plus additional conditions) |
| Updated | Evidence collected 26 Sept 2026 |
How the rating was worked out
Weights for Orchestrators & meta-harnesses: adoption 40%, experts 30%, crowd 30%. Fewer than two pillars have enough evidence. The method.
| Adoption & momentumweight 40% | 88 | GitHub stars 51k → 82 · 61 releases in 90 days → 100reach × 0.7 + momentum × 0.3 |
|---|---|---|
| Expertsindependent reviews · weight 30% | — | Only 2 independent reviews (3 needed) |
| Crowdcommunity sample · weight 30% | — | Only 4 community items sampled (6 needed) |
| Rating | — | Not rated: fewer than two pillars have enough evidence. |
Evidence checks. 1 item removed: dated before the thread it cites (1). 1 review from tool directories, AI-written pages or unnamed authors not counted. 1 GitHub issue not counted as sentiment. How evidence is checked.
Trust & safety
Seven dimensions graded A–D from documented facts. Overall: B.
C
Permission model
Agent CLIs run as child processes with the full permissions of the OS user running the daemon; the daemon launches Claude Code with --permission-mode bypassPermissions. Docs: 'Multica does not pretend to be the boundary. Put one around it.' Isolation (dedicated user, container, VM) and review gates before merge are available but up to the operator. source
B
Data access scope
Runs inherit the daemon user's real HOME, so all files under it are reachable; per-run working directories and run-scoped API tokens exist. Member roles and per-member agent access scopes are documented. source
A
Data storage
Fully self-hostable (Docker Compose/Helm); code stays on the user's runtime. Hosted Multica Cloud stores workspaces, issues, comments, chat and attachments on AWS and Vercel. source
B
Data retention & training
Privacy policy (updated 2026-09-24): 'Multica does not use your content to train AI models.' Content is kept for as long as the account or workspace exists; deletion requests answered within 30 days. No fixed retention period. source
B
Incident history (24 months)
No CVEs or published GitHub advisories found. Minor security reports are public and some remain open: CSRF Origin-header defence-in-depth gap (#1667, Apr 2026) and credential inheritance in the daemon repo cache (#2554, May 2026). A 'security vulnerability report' issue asking where to report (#1382) was still open. source
D
Compliance
No SOC 2 / ISO 27001 attestation mentioned on the site or privacy policy; a spring 2026 review said enterprise compliance features were on the roadmap. source
B
Transparency
Source code public, but under the Multica License (Apache-2.0 plus conditions restricting hosted/embedded service use), which is not an OSI licence. Detailed per-release changelogs on GitHub (61 releases in 90 days). source
The experts
2 named independent reviews found; dated reviews from the last 12 months are scored.
- positive
If you're a solo developer running occasional agent tasks, the CLI alone might be enough
ArshTechPro (DEV Community) · 21 May 2026 · dev.to - mixed
Multica is a newer project (April 2026). The documentation is solid but the ecosystem of integrations is still growing.
AJ Awan (Flowtivity) · 25 Apr 2026 · flowtivity.ai
The crowd
4 coded posts from x, github.
1 positive1 mixed2 negative
- positive
Multica is super cool - the one I've enjoyed picking up
@Trader_XO · 9 Aug 2026 · x.com - negative
After upgrading Multica to v0.4.27, I noticed several UI/loading-related issues.
multica-ai/multica discussion #7089 · 17 Aug 2026 · github.com - negative
每一次都会有100m大小，这个大小和这个机制是正常的吗？
multica-ai/multica discussion #6709 · 10 Aug 2026 · github.com - mixed
虽然可用，但是无法开启worktree模式进行任务并行，只能顺序执行，速度很慢
multica-ai/multica discussion #8571 · 19 Sept 2026 · github.com
Pricing
As published, checked 26 Sept 2026.
| Plan | Price | Notes |
|---|---|---|
| Self-hosted / internal use | Free | Site: 'Using Multica inside your own organization is free, including self-hosting it for your whole team.' |
| Commercial licence (offering Multica as a hosted service) | Custom | Required to host or embed Multica as a service for others; contact sales. |
Recent changes
- 24 Sept 2026
v0.5.3: issues auto-complete when every linked PR merges. source
- 24 Sept 2026
Privacy policy updated (no training on content; AWS/Vercel hosting; subprocessors listed). source
- 21 Sept 2026
v0.5.1: CLI supports self-hosted Gitea release sources. source
- 18 Sept 2026
v0.5.0 released (includes Hermes agent shutdown fix). source
- 7 Sept 2026
v0.4.41: task history exposes per-agent usage. source
Ranked alongside
Other ranked entries in Orchestrators & meta-harnesses.
OrcaStably AI78
SupersetSuperset Inc.70
HerdrHerdr (Can Celik; YC F26)68
T3 CodePing Labs (ping.gg / T3 Tools)67
Vibe KanbanBloop (company shut down April 2026; now community maintained)60
Sources
Every figure above traces to one of these, observed up to 26 Sept 2026.