v0.8.0 · invite-only early access

checking local daemon…

A PM for your AI coding agents. Delegate a task, watch it run, get a plain-English report.

A trust score on every file it changed. Every action recorded, every report fact-checked. Local-first. Open source.

codehere

scroll

01 · Orchestrate

Run any agent.
Catch them when they lie.

Claude Code, Codex, Aider, OpenCode, Gemini. All running in parallel, mediated by one local daemon. Every tool call audited. Every file scored. Every claim verified.

codehere · conductor

4 active

Claude Code✓ done

build rate-limit middleware

src/middleware/rate-limit.ts

Codex✓ done

adversarial security audit

src/middleware/rate-limit.ts

OpenCode78%

verifier · plan vs code

scripts/verify.ts

Aider42%

write ops runbook

docs/RATE_LIMIT.md

02 · Verify

The verifier reads what actually shipped.

Agents round to optimism. "Tests pass" means 86 ran plus 1 skipped. Codehere compares the claim against the audit chain and surfaces every gap before push.

The green check is built from things the agent can't argue with: the diff on disk, a second model's review, the hash-chained audit.

codehere verify

2 gaps

agent claimed

claude-code: "18 of 18 tests passed · 87 green ✓"

verifier found

⚠ 86 ran · 1 skipped (cache.skip.test.ts)

⚠ IPv6 test:158 is duplicate (should be 2001:db8::)

resolve

✓ auto-fix · correct Final Report before push

03 · Read your tools

Linear, Figma, draw.io as context.

Your agent reads tickets, references designs, cites diagrams. When it proposes a write back, you approve the payload before it leaves your machine.

codehere · context

3 sources

Linear

340ms

read linear://INFRA-247

API abuse on /api/* · attackers using rotating IPs

Figma

512ms

read figma://team/api/429-response

429 response shape · 4 fields · @joaquin

draw.io

280ms

read draw.io://architecture/api-edge

3 nodes · client → edge → redis

adapter contract · open · community-extensible

04 · Sync

Sync with your team.
Without a cloud.

Like Git, but for agent sessions. Signed payloads. Self-hosted remote you control. Each machine verifies independently. No central authority.

codehere · sync

chain intact

you · local

12 ahead

#0089 · 7f3a...c8de

↓signed payload · push

team-remotegit · self-hosted

git@team/codehere-chain

↓signed payload · pull · verify

teammates · local

✓ verified

#0089 · 7f3a...c8de

no central authority · each machine verifies independently

05 · FAQ

Frequently asked questions

›What is codehere?

A local-first orchestrator for AI coding agents. It mediates every agent action across Claude Code, Codex, Aider, OpenCode, Gemini, and produces a verifiable audit chain. Open source, MIT-licensed.

›How is codehere different from cloud agent workspaces?

Cloud workspaces tie your work to their servers. Codehere runs on your machine, audits every action locally, and works with the agent setup you already have. We focus on local verifiability, not cloud collaboration.

›Can I use my existing Claude Code, Codex, Aider, OpenCode, Gemini setup?

Yes. Codehere wraps the agent binaries you already have. Keeps your existing accounts, model choices, and configs. We add the safety hooks, audit chain, and verifier on top.

›Do I need to keep my laptop open?

For now, yes. Codehere runs as a local daemon. We don't tie your work to our cloud, and we don't plan to.

›How am I charged?

You're not. Codehere itself is free and open source. You pay for your own agent subscriptions. Codehere is the runtime on top.

›What about sandboxing and approvals?

Today: PreToolUse hook + optional LLM scan mediate at the tool-call boundary. Per-action approval with logged-reason override. OS-level sandbox is on the radar. Won't ship it if it breaks the easy-install promise.