Cobl

The agent connects to CRM, emails, calls and Drive to absorb the company's knowledge and deal context, and produces documents exported as PDF, Word or PowerPoint. The classes are named but the vendors are not. No named connector catalog, integrations page, custom connector or generic HTTP route is published, and which CRM and email systems it connects to, and whether it writes back to them, is not stated.

Sourcecobl.ai and the cobl.ai site mapread 2026-09-01

Cobl runs one agent in a fixed loop. The home page presents a single proactive agent that works a deal from first signal to contract in four steps (connect CRM, email, calls and drive; track CRM updates, meetings and deal strategy; produce the asset each stakeholder needs; learn from outcomes), and the security page shows it calling a fixed allowlist of twenty plus tools inside a per project sandbox with a cap on tool calls per turn. No multi-agent architecture, branching, retry, sub-workflow or customer designed sequence is documented.

Sourcecobl.ai home and securityread 2026-10-01

Documents are generated from the company's existing knowledge base and centralized, validated content, drawing on CRM data and past proposals so each output reflects the company's real expertise rather than generic templates.

Sourcecobl.airead 2026-10-01

The agent works autonomously inside the customer's workspace but stops and asks when an action carries external consequences: sharing a document publicly, sending an email or posting to a connected tool. Underneath, the agent operates against a fixed allowlist of twenty plus tools with every tool call validated against a strict schema before execution, each project runs in a dedicated sandbox the agent cannot leave, and each conversation turn is capped at a maximum number of tool calls so the agent cannot loop indefinitely.

Sourcecobl.ai/securityread 2026-10-01

Access controls are documented, and the vendor names its own attestation gap. On the Enterprise plan it offers SSO and SAML with advanced access controls, beside a dedicated sandbox per project that the agent cannot leave, AES-256 encryption, WAF and DDoS protection. The security page states that Cobl holds no independent certification today; the SOC 2 Type 2, ISO 27001 and PCI DSS certificates it lists belong to Vercel, its host, not to Cobl.

Sourcecobl.ai/securityread 2026-10-01

Outputs are kept on the record: each asset is a self contained page in a versioned repository per project, so a team can see how a document changed. No log of the agent's own steps, sources consulted or tool calls is documented, so this is a version history rather than an audit of agent behavior.

Sourcecobl.ai/securityread 2026-10-01

No per agent state carried between runs, and no memory layer with a stated scope and lifetime, is documented; what persists is stored documents. Each asset is a self-contained page in a versioned repository per project, the agent runs in a sandbox scoped to that workspace, and past documents accumulate so a customer describes tailor-made work enriching the knowledge base for reuse, which makes them content stores and business records rather than agent memory. Customer defined retention and deletion appear only in the Private Cloud and On-Premise tiers.

Sourcecobl.ai/security and cobl.airead 2026-10-01

Cobl offers three deployment models the customer chooses between, published under the heading Run Cobl your way. The default is European cloud, with all application data in Vercel's dub1 region in Dublin, product analytics through PostHog EU, and the vendor stating no data transits US infrastructure for core application operations.

Private Cloud deploys directly into the customer's own AWS or Azure tenant with native SSO and IAM and customer-defined retention and deletion rules. On-Premise runs Cobl fully inside the customer's infrastructure, keeping data in-house, integrating with their SSO and IAM, running self-hosted models with Ollama, and requiring no internet connection, with packaged updates every six months.

Sourcecobl.ai/security, Run Cobl your wayread 2026-09-01

Cobl supports multiple prebuilt document types (proposals, RFP responses, market studies, security questionnaires, ROI calculators, case studies) and lets teams capture their expertise as reusable AI agents. No broad, browsable library of packs is offered.

Sourcecobl.airead 2026-10-01

A person requests documents in the web app, though the home page calls Cobl's agents proactive, saying they know where each deal stands, surface the next move and do the work; Cobl draws on connected CRM, email, call and Drive context. No event, schedule or webhook that starts a run without a person is documented.

Sourcecobl.airead 2026-10-01

The customer chooses the provider. The security page states plainly: use Azure OpenAI, Bedrock, or your own models, and connect securely to your preferred AI providers. The on-premise model goes further, documenting self-hosted models run with Ollama with no internet connection required, so a buyer can keep inference entirely inside their own estate.

Sourcecobl.ai/security, Private Cloud and On-Premise sectionsread 2026-09-01

The site footer links API documentation at app.cobl.ai/api/docs, a first party statement that an API exists. That page is not openly readable, so its endpoints, authentication and scope are unknown.

Sourceapp.cobl.ai/api/docsread 2026-10-01

The home page states that every outcome feeds back, what worked, what did not and who converted, so Cobl gets sharper. Nothing says what feeds the loop, what it tunes or how a change is measured, and no customer test or evaluation surface is documented.

Sourcecobl.airead 2026-10-01

No browser automation or general computer use capability is documented; Cobl operates over documents, CRM data and connected files.

Sourcecobl.airead 2026-10-01