cd /news/ai-tools/end-ai-agent-lock-in-start-conversat… · home topics ai-tools article
[ARTICLE · art-119140] src=github.com ↗ pub= topic=ai-tools verified=true sentiment=↑ positive

End AI agent lock-in. Start conversation with one agent – continue with another

Aplexica, a new local open-source application, enables cross-agent state portability for AI coding agents, allowing users to switch between agents like Claude Code and Codex without losing memories, skills, tools, or conversation history. The tool replicates agent state deterministically and supports synchronization across devices, with installation via Homebrew and optional remote plugins.

read10 min views1 publishedSep 2, 2026
End AI agent lock-in. Start conversation with one agent – continue with another
Image: Michielbdejong (auto-discovered)

Cross-agent state portability for AI coding agents. Your memories, skills, tools, and conversation history — portable across every AI coding agent you use. Switch agents without starting over. Run several side-by-side, all reading from the same context. Try the next agent the day it ships and lose nothing if you don't like it.

Deterministic lossless replication of agent state — not LLM-summarized briefing.The next agent doesn't readaboutyour previous session; itisthe previous session, continued.

Aplexica is complete and usable as a local open-source application. It does not require an account, a hosted service, or a network connection. Optional remote plugins are separate, user-selected integrations and are never required for local operation.

For cross-device synchronization across your AI agents, visit aplexica.com to learn more.

brew install Aplexica/tap/aplexica
aplexica setup --yes --install

Uninstall Test. A portability tool should survive its own removal. See the Uninstall Test.

Today, picking an AI agent is a one-way door. The moment you commit to Claude Code or Codex or any other, you accumulate context that only that agent can read — memories about you, skills you've installed, MCP servers and subagents you've wired up, conversations you've built up. After a few weeks of real use, that context is what makes the agent useful. Switching means starting over. Running two in parallel means maintaining two siloed copies by hand. Trying a new agent that just launched means abandoning everything you've built.

Aplexica eliminates that lock-in. It sits beside your AI agents and turns the artifacts they produce — memories, skills, tools, and conversation history — into portable, agent-agnostic state that you own and that follows you to every agent you use.

With Aplexica installed you can:

Use every agent like it has always known you. Memories, skills, and conversation history sync among all your installed agents within seconds.Switch agents without losing a thing. One command converts your full state from any V1 agent into any other agent's native format.Fork a conversation across agents. Start in one agent, branch into a second to try a different direction, keep both. Compare results.Back up and restore. Snapshot everything your agents know, restore it after a lost laptop or OS reinstall.

After Aplexica is installed and the daemon is running, try this quick conversation sync check:

Open the local Aplexica web UI and go to

Routing rules. - Choose

Add from preset and create theSync everything everywhere routing rule template. - Open Codex and ask:

What is the largest planet in our solar system?

Wait for Codex to answer.

Open another terminal window, run

claude

, then type/resume

. - In Claude Code's conversation list, select the conversation named like:

Codex: what is the largest planet in our solar system?

Claude Code should resume the conversation that started in Codex, using its own native conversation history. For Hermes, you may need to close Hermes and open it again before the refreshed conversation list appears.

The same codex

and claude-code

adapters also serve their installed desktop apps. After writing a Codex rollout, Aplexica asks Codex's own app-server to load it; when that best-effort registration succeeds, Codex's normal thread inventory can discover it without Aplexica touching private app state. Claude Code Desktop shares configuration and project state, including guarded mirrors for its default active worktrees, but Anthropic keeps its Desktop conversation list separate. On macOS and Windows, resume the synchronized artifact in the CLI and run /desktop

for the supported app handoff.

The CLI and desktop surfaces are detected independently; either can be used by itself. The daemon retains candidate roots for these two adapters and rechecks surface availability at runtime, so installing the CLI or desktop app later activates synchronization without requiring the other surface to be present. Activation completes the native safety snapshot before first synchronization and backfills recent peer-device conversations using the configured history limit without synthesizing duplicates of locally authored sessions.

Aplexica installs three executables together:

aplexica

— CLI, daemon, local web UI server, and setup wizard.aplexica-status

— status helper the tray spawns, so process monitors can tell the watcher apart from the daemon.aplexicatray

— menu bar / system-tray companion.

The local web UI is compiled into aplexica

itself in release builds; there is no separate UI package to install.

macOS / Linux — Homebrew: Homebrew is a live channel at 1.0.74. Full steps: brew.md.

brew install Aplexica/tap/aplexica
aplexica setup --yes --install

Debian / Ubuntu — .deb: follow the exact-version download and verification steps in

apt.md, then:

aplexica setup --yes --install

Windows (release archive): download aplexica-1.0.74-windows-amd64.zip

or aplexica-1.0.74-windows-arm64.zip

from the v1.0.74 release, verify it following verify.md, unzip it into a folder you own, and run aplexica.exe

or aplexicatray.exe

from that folder. Full steps, including start-at-logon: windows.md. Then:

aplexica setup --yes --install

That one setup

command registers and starts the daemon, installs and launches the tray when it is enabled (the default on every supported desktop OS), and brings up the local web UI. Adapters are built in and auto-discover your agents.

Platform Channel Status and guide
Debian / Ubuntu Versioned .deb
Available;

direct.mdbuild.mdbrew.mdwindows.mdupdate.mdAplexica release authority is a non-exportable AWS KMS key—not GitHub, a CDN, a package registry, or a maintainer workstation. Each release ships a KMS-backed cosign signature over SHA256SUMS

and a KMS-backed, public-policy-checked SLSA v1 provenance bundle. GitHub Actions receives only a short-lived AWS session for the isolated signing job; the publication job has no signing authority.

Verify SHA256SUMS

with the independently distributed aplexica-release.pub

, check the selected artifact digest, and verify its provenance and semantic policy. GitHub Releases, a NAS, USB, or a local HTTP directory may all transport the same authenticated bytes. The exact commands are in Verify a release. Never pipe downloaded text to a shell or PowerShell expression.

Development builds use the public Go source directly:

git clone https://github.com/Aplexica/Aplexica.git
cd Aplexica
make build
make tray

The local web UI is replaced by an explanatory placeholder unless internal/web/embed/dist-local/

contains a Portal bundle. Release-style builds add -tags release

, which makes the daemon fail closed when that bundle is missing; development builds can omit the tag. See the source-build guide for tests, cross-compilation, and local installation.


aplexica setup --yes --install
aplexica status

aplexica list                                # nothing yet — agent state hasn't been imported

aplexica sync enable --all                   # or name specific agents
aplexica daemon reload


aplexica list                                # see every artifact across kinds
aplexica show <artifact-id>                  # inspect one
aplexica conflicts list                      # any divergent writes that need your decision
aplexica branch list <artifact-id>           # branch topology if you've forked

A guided walkthrough from install to working two-agent sync lives at docs/quickstart.md.

Five V1 agents, four artifact kinds, full bidirectional translation:

Agent Surfaces Vendor Memory Skill Conversation Tool Canonical conv (acf.conversation.v1 )
Claude Code CLI + Desktop Anthropic ✅ markdown ✅ skill.md ✅ session.jsonl ✅ MCP (.mcp.json)
Codex CLI + Desktop OpenAI ✅ AGENTS.md ✅ skill.md ✅ session.jsonl ✅ MCP (config.toml)
Hermes CLI Nous Research ✅ MEMORY.md / USER.md ✅ skill.md ✅ SQLite (state.db) ✅ MCP (config.yaml)
OpenClaw CLI community ✅ MEMORY/AGENTS/CLAUDE/DREAMS.md + memory/YYYY-MM-DD.md ✅ skill.md ✅ session JSONL ✅ MCP (openclaw.json)
Kilo Code CLI Kilo ✅ AGENTS.md / AGENT.md ✅ skill.md ✅ SQLite (kilo.db ) + kilo import
✅ MCP (kilo.jsonc)

Adding a new agent is one ~80-line adapter package. See docs/adapters/ for per-agent specs and internal/adapter/openclaw/ for a reference implementation.

Each supported agent has an adapter that translates between its native format and the Aplexica Canonical Format (ACF) — an append-only event log per artifact, hash-chained for integrity, plus a typed payload. State lives in the user-owned canonical store at ~/.aplexica/store/

. The daemon, run through aplexica daemon

, watches each agent's native location and runs the import → fan-out cycle on every change. Secrets live in a separate protected store and never enter the canonical event log. The whole stack is deterministic and lossless — no LLM summarization, no consolidation, no briefing layer.

The public BRDs in docs/ describe the canonical format and local product requirements.

Power-user workflows are first-class:

Branches.aplexica fork <artifact> --from <event> --to-agent codex

creates a divergent branch; the source agent's view is unchanged.aplexica log --graph <artifact>

renders the topology.aplexica merge <artifact> --from <branch> --into <branch>

reconciles.Selective sync.aplexica tag add <artifact> work

then a TOML rule likeroute.agents = ["claude-code", "codex"]

routes that artifact only to your work agents.aplexica rules test <artifact>

explains every routing decision.Stays-local content. Tags are metadata, not policy, and no routing rules are active by default. To keep aprivate

memory on its originating agent, add an explicit rule matching that tag withroute.agents = ["__originatingAgent__"]

androute.remote = "exclude"

; thequickstartgives the complete rule and safe replacement sequence.

Full reference: docs/04-brd-forking-and-merging.md, docs/05-brd-selective-sync-and-routing.md.

aplexicatray

is a cross-platform menubar / system-tray / SNI indicator that reflects daemon state at a glance:

Icon Meaning
Idle Daemon up, no recent activity
Active Snapshot ticks arriving
d Daemon up but quiet for ≥ 5 minutes
Conflict Unresolved conflicts (overrides other states)
Error Daemon not reachable

Right-click for / Resume

, Show Conflicts

, Open Logs

, Pending Projects

, and Open Aplexica

. Packaged installs include it, and aplexica daemon install

wires it to start with Aplexica on every desktop OS unless you explicitly opt out.

Path Purpose

docs/user-guide.mddocs/quickstart.mddocs/install/docs/01-brd-backup-restore.mddocs/02-brd-format-adapters.mddocs/03-brd-local-realtime-sync.mddocs/04-brd-forking-and-merging.mddocs/05-brd-selective-sync-and-routing.mddocs/09-security-and-trust-model.mddocs/10-non-functional-requirements.mddocs/adapters/docs/plugin-protocol-spec.mddocs/plugin-author-guide.mdCHANGELOG.mdAplexica welcomes contributions. Start with CONTRIBUTING.md for branch / commit / test conventions, then open a PR.

  • Contributions are accepted under a Contributor License Agreement. The CLA grants Aplexica the right to distribute your contribution under AGPL-3.0andunder a separate commercial license. A bot comments on your first PR with the signing link; seeCONTRIBUTING.mdfor the full terms. - Bug reports → issueswith thebug_report

template. - Security vulnerabilities → do not file a public issue; seeSECURITY.mdfor the responsible-disclosure address and policy. - Community conduct → CODE_OF_CONDUCT.md(Contributor Covenant 2.1).

Aplexica OSS is licensed under the GNU Affero General Public License v3.0 (LICENSE).

The AGPL-3.0 plugin-boundary stance is documented in LICENSE-EXCEPTIONS.md: third-party plugins running in their own process and communicating with the aplexica

daemon via the documented JSON-RPC stdio protocol are NOT considered derivative works of the daemon for AGPL-3.0 purposes. See that document for the full legal statement.

Aplexica is under active development. Use the release page and CHANGELOG.md for shipped-version status.

── more in #ai-tools 4 stories · sorted by recency
── more on @aplexica 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/end-ai-agent-lock-in…] indexed:0 read:10min 2026-09-02 ·