Tools
Aionis Openclaw
By aionis-openclaw-plugin connects Aionis to OpenClaw, enabling agents to record real actions and replay entire workflows instead of relying only on conversation memory.
Install
openclaw plugins install @aionis/openclawREADME
# OpenClaw Aionis Memory Plugin
[](https://www.npmjs.com/package/@aionis/openclaw)
[](https://www.npmjs.com/package/@aionis/openclaw)
[](https://github.com/Cognary/aionis-openclaw-plugin/releases)
[](https://github.com/Cognary/aionis-openclaw-plugin/blob/main/LICENSE)
Website: https://cognary.github.io/aionis-openclaw-plugin/
NPM: https://www.npmjs.com/package/@aionis/openclaw
Releases: https://github.com/Cognary/aionis-openclaw-plugin/releases
Most agent memory plugins only store chat history.
Aionis goes further.
It turns memory into an execution loop:
`Memory -> Policy -> Action -> Replay`
Your agent does not just remember.
It learns how to act better over time.
Aionis is production memory infrastructure for agents.
This plugin brings Aionis into OpenClaw (including Clawbot) and turns memory into an executable loop:
`Memory -> Policy -> Action -> Replay`
## What's new in v0.1.2
- Replay execution integration expanded for Clawbot workflows and repair-review controls.
- Replay selfcheck improved with strict/guided mode checks and stronger status gating.
- Local standalone bootstrap defaults aligned to Aionis `standalone-v0.2.7`.
- Public landing page added and deployed on GitHub Pages.
## Why Aionis
Most memory stacks stop at retrieval. Production teams still need:
- auditable and replayable writes
- reliable write paths decoupled from embedding availability
- policy loops that can influence runtime behavior
- operational guardrails for long-running systems
Aionis is designed as a long-running memory kernel to address those requirements.
## Core strengths
1. Verifiable write chain
- Every write can be traced with `commit_id` and `commit_uri`.
2. URI-first object model
- Nodes, edges, commits, and decisions are referenceable via stable URIs across API/SDK/Ops.
3. Layered context orchestration
- Context is assembled in explicit layers: `facts / episodes / rules / decisions / tools / citations`, with budget controls.
4. Memory to policy loop
- Memory can affect routing and execution using `tools/select` and `tools/feedback`.
5. Production evidence and gates
- Readiness is validated through reproducible checks, runbooks, and benchmark artifacts.
6. Sandbox interfaces (experimental in Aionis core)
- Controlled sandbox APIs can be linked to policy/replay lineage.
## What this plugin does
- Auto recall on `before_agent_start` with `POST /v1/memory/context/assemble`
- Auto capture on `agent_end` with `POST /v1/memory/write`
- Auto policy feedback on successful turns with `POST /v1/memory/tools/feedback`
- Exposes policy actions: `POST /v1/memory/tools/select`, `POST /v1/memory/tools/feedback`
- Exposes replay actions:
- record: `POST /v1/memory/replay/run/start`, `step/before`, `step/after`, `run/end`
- query/compile: `POST /v1/memory/replay/runs/get`, `playbooks/compile_from_run`, `playbooks/get`
- lifecycle/run: `POST /v1/memory/replay/playbooks/promote`, `playbooks/repair`, `playbooks/repair/review`, `playbooks/run`
- Manual tools:
- memory/policy: `aionis_memory_search`, `aionis_memory_store`, `aionis_memory_context`, `aionis_policy_select`, `aionis_policy_feedback`
- replay: `aionis_replay_run_start`, `aionis_replay_step_before`, `aionis_replay_step_after`, `aionis_replay_run_end`, `aionis_replay_run_get`, `aionis_replay_playbook_compile`, `aionis_replay_playbook_get`, `aionis_replay_playbook_promote`, `aionis_replay_playbook_repair`, `aionis_replay_playbook_repair_review`, `aionis_replay_playbook_run`
## 30-second setup
```bash
openclaw plugins install @aionis/openclaw && \
openclaw aionis-memory bootstrap && \
openclaw aionis-memory selfcheck --scope clawbot:selfcheck
```
Replay path selfcheck (all modes):
```bash
openclaw aionis-memory replay-selfcheck --scope clawbot:selfcheck --mode simulate
openclaw aionis-memory replay-selfcheck --scope clawbot:selfcheck --mode strict --backend local_process
openclaw aionis-memory replay-selfcheck --scope clawbot:selfcheck --mode guided --backend sandbox_sync --project-id clawbot-demo
```
## Project isolation in Clawbot
This plugin isolates memory by `tenant_id + scope`.
Default mode is `scopeMode=project`:
- It derives workspace path from OpenClaw context (`workspace/cwd`).
- It generates project scope as: `scopePrefix + ":" + basename(workspacePath) + "-" + sha1(workspacePath)[0:8]`.
- All write/recall/policy calls use that scope, so different projects do not mix memories.
Other modes:
- `fixed`: one global scope
- `session`: one scope per session
- `project`: one scope per workspace/repo (recommended)
## Security and contract
- Tenant isolation: `tenant_id + scope`
- Memory auth: API key and bearer-token modes in Aionis core
- Admin interface: separate token channel in Aionis core
- Public API contracts: stable `/v1/memory/*` shapes documented in Aionis docs
## Quick start (local standalone)
1. Start local Aionis standalone:
```bash
./bootstrap-local-standalone.sh
```
Default image: `ghcr.io/cognary/aionis:standalone-v0.2.7`.
If `3001` is occupied, the script auto-falls back to `3002-3010`.
You can also force a port:
```bash
AIONIS_PORT=3002 ./bootstrap-local-standalone.sh
```
This script writes:
- `~/.openclaw/plugins/aionis/aionis.env`
- `~/.openclaw/plugins/aionis/clawbot.env`
2. Install plugin (from npm):
```bash
openclaw plugins install @aionis/openclaw
```
3. Bootstrap plugin config automatically:
```bash
openclaw aionis-memory bootstrap
```
4. Verify end-to-end path (recommended sequence):
```bash
openclaw aionis-memory selfcheck --scope clawbot:selfcheck
openclaw aionis-memory replay-selfcheck --scope clawbot:selfcheck --mode simulate
openclaw aionis-memory replay-selfcheck --scope clawbot:selfcheck --mode strict --backend local_process
openclaw aionis-memory replay-selfcheck --scope clawbot:selfcheck --mode guided --backend sandbox_sync --project-id clawbot-demo
```
## Configuration
- `baseUrl`: Aionis API base URL
- `apiKey`: API key (`x-api-key`)
- `tenantId`: tenant id
- `scope`: default fallback scope
- `scopeMode`: `fixed`, `session`, `project`
- `scopePrefix`: prefix for derived scopes
- `preset`: `compact` (default), `policy-first`, `custom`
- `autoRecall`: inject memory context before each turn
- `autoCapture`: persist dialogue summary after successful turn
- `autoPolicyFeedback`: write `tools/feedback` after successful turn
- `includeShadow`: include shadow rules in policy calls
- `strictTools`: strict mode for `tools/select`
- `recallLimit`: recall/context limit
- `captureMessageLimit`: max recent messages captured
- `contextCharBudget`: max injected context size
- `debug`: verbose logs
## Presets
- `compact` (default): lower token budget and lower recall fan-out
- `policy-first`: larger context and stricter routing defaults
- `custom`: disable preset defaults and use explicit values
## Troubleshooting
- `404 Route ... not found`: verify Aionis version and `baseUrl`
- `400 invalid_request`: verify payload and required fields
- `strict/guided replay run failed`: check sandbox settings (`SANDBOX_ENABLED=true`, local executor policy, command allowlist), and verify `allow_local_exec` (plugin selfcheck auto-enables for strict/guided)
- Policy feedback logs include: `policy switch reduced` or `policy switch detected`, and `rule confidence updated (updated_rules=...)`
## License
Apache License 2.0. See [LICENSE](./LICENSE).
## Release
```bash
npm run -s release:preflight
npm run -s release:tgz
```
Reference:
- [RELEASE_CHECKLIST.md](./RELEASE_CHECKLIST.md)
- [USAGE_FULL.md](./USAGE_FULL.md)
tools
Comments
Sign in to leave a comment