Comp AI CRM

Self-hosted CRM where a scheduled research agent keeps the records, tuned by four markdown skill files.

AdaptableOperable Business appWeb app Hybrid TypeScript Screen only License: MIT

An MIT-licensed CRM whose built-in agent runs on its own schedule, reads your mail threads and signature blocks, researches contacts, and writes only strongly evidenced facts back. Three deployments (Next.js app, NestJS tRPC API, agent) share one .env; the agent's behavior lives in prompt files you can edit. There is no token-authenticated API for an outside agent yet.

Use it when

I want a CRM that fills itself in from our inboxes instead of nagging reps to log calls
I want to change what the research agent is allowed to look at and write, without rewriting the app

Gotchas

No API tokens: every request needs an identity-provider session, so an outside agent has no documented way in
Three separate deployments must agree on DATABASE_URL and BETTER_AUTH_SECRET or you get a redirect loop
Microsoft client secrets expire (6 months default) and mail sync fails silently when they do
Default branch is release, not main

Last activity Sep 2, 2026. Fit: adaptable 3, operable 1.

Editions

Free · Free open source

Clonegit clone https://github.com/trycompai/crm
ImageContainer

For agents: JSON · Markdown

Adaptable: what is yours to change

Make it yours

Config only27 things to edit

One .env at the repo root feeds all three processes; the README table lists which values matter. Sign-in is an allow-list (ALLOWED_SIGN_IN) plus Google or Microsoft OAuth. The agent's judgement is prose in apps/agent/agent/skills/*.md and instructions.md, versioned like code; edit those to change what counts as evidence, how identities are matched, and what a brief looks like. Optional keys open more sources.

Env vars: DATABASE_URL BETTER_AUTH_SECRET ALLOWED_SIGN_IN GOOGLE_CLIENT_ID GOOGLE_CLIENT_SECRET MICROSOFT_CLIENT_ID MICROSOFT_CLIENT_SECRET MICROSOFT_TENANT_ID API_URL APP_URL AUTH_COOKIE_DOMAIN PERPLEXITY_API_KEY AI_GATEWAY_API_KEY AGENT_BRIDGE_SECRET REDIS_URL CRON_SECRET SLACK_CLIENT_ID SLACK_CLIENT_SECRET CRM_TELEMETRY_DISABLED
Config files: .env.example docker-compose.yml apps/agent/agent/schedules/dispatch.ts
Prompt files: apps/agent/agent/instructions.md apps/agent/agent/skills/evidence.md apps/agent/agent/skills/identity-matching.md apps/agent/agent/skills/data-boundaries.md apps/agent/agent/skills/writing-a-brief.md
Accounts you need: Google Cloud OAuth client or Microsoft Entra app (at least one)Perplexity (optional)Context (optional)Vercel AI Gateway (optional)

Source

github.com/trycompai/crm
MIT license
Last commit Sep 2, 2026
9,383 stars

Operable: what an agent can drive

Interface

GUIScreen onlypublished by the builderlocal

How an agent can drive an Operable Listing, in five levels: native (built for the agent as the primary interface), official (the Builder ships an agent interface), wrapped (a third party wraps the software's API), config (an agent can only drive it by editing files), gui (screen control only). An interface is official only when it is published by the same organization that ships the software. GUI-level Listings are allowed and sorted last.

Docs: github.com/trycompai/crm
Tools: read_crm_history search_crm identify_contact research_person enrich_company record_fact schedule_recheck query_crm read_crm_record create_crm_activity post_slack_message

Reach

Partial coverage. 2 of 7 jobs reachable through the stated interface.

Job	Through the interface	Evidence
Research and enrich a contact or company	Not reachable	Runs on the bundled agent's schedule; no endpoint or token for an outside agent to trigger it
Read pipeline records	Not reachable	tRPC router in apps/api sits behind a Better Auth IdP session cookie; README: 'there is no way to get a session from a terminal'
Write facts and activities to a record	Not reachable	Same tRPC surface, same cookie-only auth; no API token or key flow documented
Talk to the agent about a contact	Not reachable	Agent tab in the web app, guarded by AGENT_BRIDGE_SECRET; not an external endpoint
Sync Gmail or Microsoft mailboxes	Reachable	POST /internal/sync/mailboxes guarded by the CRON_SECRET env var, documented in README Deploying; the one non-interactive endpoint
Schedule a re-check on a person	Not reachable	schedule_recheck is an internal agent tool, not exposed outside the agent
Manage sign-in allow-list and SSO	Reachable	ALLOWED_SIGN_IN env var; Settings → SSO in-app

Code stack

Languages: TypeScript
Frameworks: Next.jsNestJSnestjs-trpcPrismaBetter Autheve
Package manager: bun
Database: PostgreSQLRedis (optional)
Hosting target: VercelDocker
Built: Unknown (recorded, never ranked)

Model stack

Providers: Vercel AI GatewayPerplexity (optional)Context (optional)
Models: none
Keys you need: AI_GATEWAY_API_KEY PERPLEXITY_API_KEY AGENT_BRIDGE_SECRET
Prompts live in: apps/agent/agent/instructions.md apps/agent/agent/skills/evidence.md apps/agent/agent/skills/identity-matching.md apps/agent/agent/skills/data-boundaries.md apps/agent/agent/skills/writing-a-brief.md apps/agent/agent/subagents/agent_runner/instructions.md

Skills

One repo tooling skill for AI UI elements used by contributors.

Kind: contributor
Paths: .agents/skills/ai-elements/SKILL.md

Evidence and confidence

Field	Where it was seen	Confidence
formFactor	README: Next.js app plus NestJS API plus agent deployment	95%
executionModel	README The agent: model-driven research with a fixed dispatcher and evidence ledger; runs with no keys at all	90%
operability	README (release branch): app talks to the API over tRPC; sign-in only through Google/Microsoft IdP, session cookie minted by Better Auth; no token, API key, or REST docs. Only non-interactive route is POST /internal/sync/mailboxes with CRON_SECRET. Screen control of the web app is the fallback.	90%
effort	README Configuration table and .env.example; skills are markdown files in apps/agent/agent/skills	90%
editions	README License: MIT; no paid edition found	80%
active	GitHub pushedAt 2026-09-02	–
reach		80%

Open questions

Is a token-authenticated API or MCP endpoint planned for outside agents?
Which model does the agent call through AI_GATEWAY_API_KEY by default?
Is trycrm.ai a hosted edition with pricing?

Sources

github.com/trycompai/crm sourcing:smb-ops-crm:github
trycrm.ai sourcing:smb-ops-crm:github