Comp AI's agentic CRM, or an approval boundary the agent cannot cross?

Where Comp AI's CRM wins

It proved the demand is real. The research file this page reads — docs/strategy/COMPETITOR_MAP.md, dated August 4, 2026 — records 4.4k GitHub stars gathered in weeks for an agentic-first CRM: an autonomous background agent with a work queue, evidence-based facts and a sandboxed design the file calls genuinely novel, under an MIT licence, TypeScript end-to-end on Next.js, NestJS and Prisma. If what you want is a CRM in which the agent is a worker inside the product, that is its entire thesis — and nothing in this framework attempts it. There is no background agent here, no work queue, and no scheduler for either: nothing in this framework runs unattended at all (L-04).

It is also far simpler to stand up. The file records deployment as three Vercel deployments. This project is not a product at all — no hosted CRM, no free tier, no account (L-07) — with no authentication, tenancy or RBAC (L-01), persistence in local SQLite (L-02), and a bootstrap that runs from a checkout of the repository rather than from a registry, so ownership means copying source and upgrading means merging (L-08).

Where this is different

The direction of the agent. The file's sentence is exact: the agent is in the product, not building it. Here it is the reverse. The application is generated as code in your repository — a module manifest becomes a migration, a service, a REST resource, an SDK method and Admin screens (C-01) — and the runtime surface an agent gets is refusal-shaped rather than autonomous. Commercial policy is deterministic code, and a renewal at or above the threshold stops and waits for a named human (C-03). An agent actor asking to approve a discounted quote is refused with a 403 and code HUMAN_APPROVAL_REQUIRED, and only a human user actor can decide (C-21). The refusal is asserted by a test, so the boundary is a property of the system rather than a promise in a README (C-04), and every mutation leaves an audit event and a step-level trace behind it (C-16). Autonomy is not the pitch here; the approval boundary is.

Ownership shape. The file records Comp AI's customization as none meaningful — fixed schema, single-tenant, Vercel-coupled — with no MCP and 18 internal agent tools. Here the schema is whatever your manifests declare, generated into source you own and evolve through explicit revisions (C-01), and the lock-in the file records for this framework is none by design — a framework as dependency, though today that means copying source rather than installing one (L-08).

The measurement caveat cuts both ways and belongs on this page. The file records 81 commits behind those 4.4k stars and calls the project a manifesto, not a platform — but the same file warns that stars in this category are marketing-inflected, citing Krayin at 23.6k stars with modest activity, and says to measure forks, usage and benchmarks instead. Held to its own standard, this project's benchmark has not been run, so no measured comparison exists in either direction and none appears here (L-03).

How to choose

Choose Comp AI's CRM if you want an autonomous CRM worker inside a product today; if MIT licensing, a TypeScript-end-to-end stack and Vercel deployment fit; and if the fixed schema and single-tenant shape the file records are acceptable for what you are running.

Choose this framework in the narrow case: the deliverable is a bespoke commercial process as reviewable code in your repository, where the agent does the writing and a human owns every approval — and you can supply authentication, hosting, PostgreSQL and data governance yourself, because none of them is here (L-01, L-02).

If the question is which agent should touch revenue decisions, the two answers differ in kind rather than degree: theirs works the queue, and this one is stopped at the approval and told to wait for a person. Choose by which failure you would rather explain afterwards.

What this comparison does not cover

No hands-on evaluation. Nothing from Comp AI was installed, configured, benchmarked or run in producing this page. Every fact traces to docs/strategy/COMPETITOR_MAP.md, researched on August 4, 2026, and this environment has no network access to re-verify any of it.

No review of the evidence model or the sandbox. The file records the design as genuinely novel and does not describe how it works, so this page cannot and does not.

No feature matrix and no pricing. The file carries neither for Comp AI, so no statement about its CRM features, data model or cost appears here.

The commit and star figures are a snapshot from one date, in a category where the file itself says stars are marketing-inflected. Check the current repository before drawing conclusions from either number.

No build-rate or speed comparison in either direction — the benchmark has not been run (L-03). When the competitor research is refreshed, this entry must be re-read against it.