Early access · onboarding design partners

Claude agents that do real work for your team.

Cofedev builds two products on Claude. Nexilume runs teams of agents for support and operations. Revuo reviews every pull request with a panel of AI reviewers.

Become a design partner See the products
nexilume · support-agent · TelegramNexilume
  1. contextLoaded tenant memory + 3 knowledge docs41ms
  2. thinkCustomer wants to move a booking date1.2s
  3. toolsupport_case.lookup("A-2291")180ms
  4. observeChange allowed until 48h before the date12ms
  5. thinkDraft reply, cite policy 4.20.9s
  6. grounding2 claims · 2 cited · 0 unsupported64ms
  7. finalizeReply sent · case updated · memory saved22ms
Example runIllustrative timings
api/orders/refund.ts · PR #148Revuo
41  const order = await getOrder(id)
42- const amount = req.body.amount
42+ const amount = order.total
43  await refund(order, amount)
revuo · security reviewer ● blocker The refund amount came from the request body, so a client could refund more than it paid. Using the stored order total closes that path.
Example review6 reviewers · 1 verified finding
Nexilume · agent operations platform

Your operations, run by teams of Claude agents.

A multi-tenant platform to build, run and govern agent teams. Agents answer customers where they already are, follow workflows you define, and keep long-term memory of your business.

Teams that delegate

Shared task boards, handoffs, subagents and peer review.

Workflows as data

Routing and flows defined per tenant, with no custom code.

Memory that compounds

Session summaries, a knowledge graph and a document vault.

Every channel

Telegram, WhatsApp, Slack, Discord, LINE, Zalo, Lark and APIs.

Code workspaces

Isolated git worktrees with checkpoints, test runs and previews.

Tools, skills, MCP

Built-in tools, MCP servers, skill packages and schedules.

Stages that loop until the task is done
ContextLoad memory and tenant knowledge
PruneKeep the context focused
ThinkClaude plans the next step
ToolAct through permitted tools
ObserveRead results, checkpoint
GroundingCheck claims against evidence
FinalizeReply, persist, remember
Tenant isolationRole-based accessAES-256-GCM key encryptionSandboxed executionSSRF protectionAudit trailsSelf-hostable
Revuo · AI code review

Six reviewers on every pull request.

Revuo reviews pull requests and merge requests on GitHub and GitLab. Each persona reads the diff in parallel, a verifier drops weak findings, and the rest land as inline comments with a verdict.

GeneralSecurityPerformanceArchitectureBusiness logicAccuracy

Your team's rules

Custom rules on top of rule sets for 100+ stacks.

Requirements coverage

Attach specs and see what each change covers.

Codebase memos

Reviews learn each repo's conventions over time.

Review analytics

Findings by severity, reviewer and repository.

Teams and plans

Repos, members and billing in one dashboard.

Model choice

Claude by default, per-reviewer models and fallbacks.

WebhookA PR opens or updates
DetectFind stacks, load matching rules
ReviewSix personas in parallel
VerifyMerge duplicates, drop weak findings
PostInline comments, summary, verdict
Built on Claude

Claude is the reasoning engine in both products.

  • Native integrationDirect Messages API client with streaming and tool use.
  • Prompt cachingSystem prompts and tenant knowledge stay cached across multi-step runs.
  • Extended thinkingRoom to reason on code changes and multi-step cases.
  • MCP and skillsThe same building blocks as Claude Code.
  • Verified outputGrounded replies in Nexilume, a verifier pass in Revuo.
What Claude credits unlock
  • Design-partner pilotsReal support, operations and review workloads for first customers, without passing model costs to them.
  • Evaluation at scaleReplay agent runs and past reviews against new Claude models before rollout.
  • Multi-agent by defaultAgent teams and six-reviewer panels use more tokens per task. Credits make them the standard setup.
Contact

We're looking for design partners.

If your team handles support, operations or code review that agents could take on, we'd like to run a pilot with you.

hello@cofedev.com